Tighten up timings a bit more

our pyproject.toml version was 0.3.68 - update to .69 in line with
release!!

.github/workflows/build-app.yml

@@ -159,7 +159,7 @@ jobs:
           fi
 
       - name: Install Homebrew packages
-        run: brew install just awscli macmon
+        run: brew install just awscli
 
       - name: Install UV
         uses: astral-sh/setup-uv@v6
@@ -243,6 +243,14 @@ jobs:
       # Build the bundle
       # ============================================================
 
+      - name: Add pinned macmon to PATH
+        run: |
+          MACMON_DIR=$(nix develop --command sh -c 'dirname $(which macmon)')
+          echo "Using macmon from: $MACMON_DIR"
+          echo "$MACMON_DIR" >> $GITHUB_PATH
+          # Remove any Homebrew macmon so PyInstaller can't accidentally pick it up
+          brew uninstall macmon 2>/dev/null || true
+
       - name: Build PyInstaller bundle
         run: uv run pyinstaller packaging/pyinstaller/exo.spec
 

.mlx_typings/mlx/core/__init__.pyi

@@ -215,6 +215,22 @@ class StreamContext:
         traceback: object | None = ...,
     ) -> None: ...
 
+def device_info() -> dict[str, str | int]:
+    """
+    Get information about the GPU device and system settings.
+
+    Currently returns:
+
+    * ``architecture``
+    * ``max_buffer_size``
+    * ``max_recommended_working_set_size``
+    * ``memory_size``
+    * ``resource_limit``
+
+    Returns:
+        dict: A dictionary with string keys and string or integer values.
+    """
+
 def abs(a: array, /, *, stream: Stream | Device | None = ...) -> array:
     """
     Element-wise absolute value.
@@ -2380,7 +2396,7 @@ def degrees(a: array, /, *, stream: Stream | Device | None = ...) -> array:
         array: The angles in degrees.
     """
 
-def depends(inputs: array | Sequence[array], dependencies: array | Sequence[array]):
+def depends[T](inputs: T, dependencies: array | Sequence[array]) -> T:
     """
     Insert dependencies between arrays in the graph. The outputs are
     identical to ``inputs`` but with dependencies on ``dependencies``.

.mlx_typings/mlx/nn/__init__.pyi

@@ -1,9 +1,5 @@
-"""
-This type stub file was generated by pyright.
-"""
-
-from layers import *
-from utils import *
+from .layers import *
+from .utils import *
 
 from . import init as init
 from . import losses as losses

.mlx_typings/mlx/nn/layers/__init__.pyi

@@ -1,20 +1,16 @@
-"""
-This type stub file was generated by pyright.
-"""
-
-from activations import *
-from base import *
-from containers import *
-from convolution import *
-from convolution_transpose import *
-from distributed import *
-from dropout import *
-from embedding import *
-from linear import *
-from normalization import *
-from pooling import *
-from positional_encoding import *
-from quantized import *
-from recurrent import *
-from transformer import *
-from upsample import *
+from .activations import *
+from .base import *
+from .containers import *
+from .convolution import *
+from .convolution_transpose import *
+from .distributed import *
+from .dropout import *
+from .embedding import *
+from .linear import *
+from .normalization import *
+from .pooling import *
+from .positional_encoding import *
+from .quantized import *
+from .recurrent import *
+from .transformer import *
+from .upsample import *

.mlx_typings/mlx/nn/layers/base.pyi

@@ -53,7 +53,7 @@ class Module(dict):
         mx.eval(model.parameters())
     """
 
-    __call__: Callable
+    def __call__(self, *args: Any, **kwargs: Any) -> mx.array: ...
     def __init__(self) -> None:
         """Should be called by the subclasses of ``Module``."""
 
@@ -200,7 +200,7 @@ class Module(dict):
     ) -> mx.MX_ARRAY_TREE:  # -> dict[Any, Any | dict[Any, Any | dict[Any, Any] | list[Any]] | dict[Any, Any] | list[Any]]:
         """Return the submodules that do not contain other modules."""
 
-    def update(self, parameters: dict, strict: bool = ...) -> Module:
+    def update(self, parameters: dict[str, Any], strict: bool = ...) -> Module:
         """Replace the parameters of this Module with the provided ones in the
         dict of dicts and lists.
 

.mlx_typings/mlx/nn/layers/convolution.pyi

@@ -32,6 +32,7 @@ class Conv1d(Module):
     """
 
     weight: mx.array
+    bias: mx.array | None
     groups: int
     def __init__(
         self,

.mlx_typings/mlx/nn/layers/linear.pyi

@@ -40,6 +40,10 @@ class Linear(Module):
         bias (bool, optional): If set to ``False`` then the layer will
           not use a bias. Default is ``True``.
     """
+
+    weight: mx.array
+    bias: mx.array | None
+
     def __init__(self, input_dims: int, output_dims: int, bias: bool = ...) -> None: ...
     def __call__(self, x: mx.array) -> mx.array: ...
     def to_quantized(

.mlx_typings/mlx/nn/layers/normalization.pyi

@@ -88,6 +88,9 @@ class RMSNorm(Module):
         dims (int): The feature dimension of the input to normalize over
         eps (float): A small additive constant for numerical stability
     """
+
+    weight: mx.array
+
     def __init__(self, dims: int, eps: float = ...) -> None: ...
     def __call__(self, x) -> mx.array: ...
 

.mlx_typings/mlx/utils.pyi

@@ -7,7 +7,10 @@ from typing import Any, Callable, Dict, List, Optional, Tuple, Union
 from mlx.core import MX_ARRAY_TREE
 
 def tree_map(
-    fn: Callable, tree: Any, *rest: Any, is_leaf: Optional[Callable] = ...
+    fn: Callable[..., Any],
+    tree: Any,
+    *rest: Any,
+    is_leaf: Callable[..., bool] | None = ...,
 ) -> Any:
     """Applies ``fn`` to the leaves of the Python tree ``tree`` and
     returns a new collection with the results.
@@ -44,11 +47,11 @@ def tree_map(
     """
 
 def tree_map_with_path(
-    fn: Callable,
+    fn: Callable[..., Any],
     tree: Any,
     *rest: Any,
-    is_leaf: Optional[Callable] = ...,
-    path: Optional[Any] = ...,
+    is_leaf: Callable[..., bool] | None = ...,
+    path: str | None = ...,
 ) -> Any:
     """Applies ``fn`` to the path and leaves of the Python tree ``tree`` and
     returns a new collection with the results.
@@ -80,9 +83,9 @@ def tree_map_with_path(
 def tree_flatten(
     tree: Any,
     prefix: str = ...,
-    is_leaf: Optional[Callable] = ...,
-    destination: Optional[Union[List[Tuple[str, Any]], Dict[str, Any]]] = ...,
-) -> Union[List[Tuple[str, Any]], Dict[str, Any]]:
+    is_leaf: Callable[..., bool] | None = ...,
+    destination: list[tuple[str, Any]] | dict[str, Any] | None = ...,
+) -> list[tuple[str, Any]] | dict[str, Any]:
     """Flattens a Python tree to a list of key, value tuples.
 
     The keys are using the dot notation to define trees of arbitrary depth and
@@ -118,7 +121,7 @@ def tree_flatten(
             the Python tree.
     """
 
-def tree_unflatten(tree: Union[List[Tuple[str, Any]], Dict[str, Any]]) -> Any:
+def tree_unflatten(tree: list[tuple[str, Any]] | dict[str, Any]) -> Any:
     """Recreate a Python tree from its flat representation.
 
     .. code-block:: python

.mlx_typings/mlx_lm/generate.pyi

@@ -30,7 +30,7 @@ def str2bool(string):  # -> bool:
 def setup_arg_parser():  # -> ArgumentParser:
     """Set up and return the argument parser."""
 
-generation_stream = ...
+generation_stream: mx.Stream
 
 @contextlib.contextmanager
 def wired_limit(
@@ -73,9 +73,11 @@ class GenerationResponse:
     finish_reason: Optional[str] = ...
 
 def maybe_quantize_kv_cache(
-    prompt_cache, quantized_kv_start, kv_group_size, kv_bits
-):  # -> None:
-    ...
+    prompt_cache: Any,
+    quantized_kv_start: int | None,
+    kv_group_size: int | None,
+    kv_bits: int | None,
+) -> None: ...
 def generate_step(
     prompt: mx.array,
     model: nn.Module,
@@ -252,48 +254,92 @@ class BatchResponse:
 
     texts: List[str]
     stats: BatchStats
+    caches: Optional[List[List[Any]]]
 
+def _left_pad_prompts(prompts: Any, max_length: Optional[int] = ...) -> mx.array: ...
+def _right_pad_prompts(prompts: Any, max_length: Optional[int] = ...) -> mx.array: ...
+def _make_cache(
+    model: Any, left_padding: Any, max_kv_size: Optional[int]
+) -> List[Any]: ...
+def _merge_caches(caches: Any) -> List[Any]: ...
 @dataclass
 class Batch:
     uids: List[int]
     y: mx.array
-    logprobs: mx.array
+    logprobs: List[mx.array] | mx.array
     max_tokens: List[int]
     num_tokens: List[int]
     cache: List[Any]
-    def __len__(self):  # -> int:
-        ...
-    def filter(self, keep_idx: List[int]):  # -> None:
-        ...
-    def extend(self, other):  # -> None:
-        ...
+    samplers: List[Callable[[mx.array], mx.array] | None]
+    logits_processors: List[List[Callable[[mx.array, mx.array], mx.array]]]
+    tokens: List[mx.array]
+    def __len__(self) -> int: ...
+    def filter(self, keep_idx: List[int]) -> None: ...
+    def extend(self, other: "Batch") -> None: ...
+    def extract_cache(self, idx: int) -> List[Any]: ...
 
 class BatchGenerator:
+    model: nn.Module
+    sampler: Callable[[mx.array], mx.array]
+    stop_tokens: set[int]
+    max_kv_size: Optional[int]
+    prefill_step_size: int
+    completion_batch_size: int
+    prefill_batch_size: int
+    unprocessed_prompts: List[Any]
+    active_batch: Optional[Batch]
+    prompt_progress_callback: Callable[[List[Tuple[int, int, int]]], None]
+    _stats: BatchStats
+    _next_count: int
+
     @dataclass
     class Response:
         uid: int
         token: int
         logprobs: mx.array
         finish_reason: Optional[str]
+        prompt_cache: Any
 
     def __init__(
         self,
-        model,
+        model: Any,
         max_tokens: int = ...,
-        stop_tokens: Optional[set] = ...,
+        stop_tokens: Optional[set[int]] = ...,
         sampler: Optional[Callable[[mx.array], mx.array]] = ...,
+        logits_processors: Optional[
+            List[Callable[[mx.array, mx.array], mx.array]]
+        ] = ...,
         completion_batch_size: int = ...,
         prefill_batch_size: int = ...,
         prefill_step_size: int = ...,
+        prompt_progress_callback: Optional[
+            Callable[[List[Tuple[int, int, int]]], None]
+        ] = ...,
+        max_kv_size: Optional[int] = ...,
     ) -> None: ...
+    def close(self) -> None: ...
     def insert(
-        self, prompts, max_tokens: Union[List[int], int, None] = ...
-    ):  # -> list[Any]:
-        ...
-    def stats(self):  # -> BatchStats:
-        ...
-    def next(self):  # -> list[Any]:
-        ...
+        self,
+        prompts: Any,
+        max_tokens: Union[List[int], int, None] = ...,
+        caches: Any = ...,
+        samplers: Optional[List[Any]] = ...,
+        logits_processors: Optional[List[Any]] = ...,
+    ) -> List[int]: ...
+    def remove(
+        self, uids: List[int], return_prompt_caches: bool = ...
+    ) -> Optional[dict[int, List[Any]]]: ...
+    def stats(self) -> BatchStats: ...
+    def next(self) -> List[Response]: ...
+    def _process_prompts(self, prompts: List[Any]) -> Batch: ...
+    def _step(
+        self,
+        input_tokens: mx.array,
+        prompt_cache: List[Any],
+        samplers: Optional[List[Any]],
+        logits_processors: Optional[List[Any]],
+        tokens: List[mx.array],
+    ) -> Tuple[mx.array, List[mx.array]]: ...
 
 def batch_generate(
     model,

.mlx_typings/mlx_lm/models/cache.pyi

@@ -16,7 +16,7 @@ class Cache(Protocol):
         self, keys: mx.array, values: mx.array
     ) -> tuple[mx.array, mx.array]: ...
     @property
-    def state(self) -> tuple[mx.array, mx.array]: ...
+    def state(self) -> tuple[mx.array | None, mx.array | None]: ...
     @state.setter
     def state(self, v) -> None: ...
 
@@ -88,17 +88,18 @@ def create_attention_mask(
 ) -> array | Literal["causal"] | None: ...
 
 class _BaseCache(Cache):
-    keys: mx.array
-    values: mx.array
+    keys: mx.array | None
+    values: mx.array | None
     offset: int
     @property
-    def state(self) -> tuple[mx.array, mx.array]: ...
+    def state(self) -> tuple[mx.array | None, mx.array | None]: ...
     @state.setter
     def state(self, v) -> None: ...
     @property
     def meta_state(self) -> Literal[""]: ...
     @meta_state.setter
     def meta_state(self, v) -> None: ...
+    def trim(self, n: int) -> int: ...
     def is_trimmable(self) -> Literal[False]: ...
     @classmethod
     def from_state(cls, state, meta_state) -> Self: ...
@@ -114,15 +115,13 @@ class ConcatenateKVCache(_BaseCache):
     def update_and_fetch(self, keys, values):  # -> tuple[Any | array, Any | array]:
         ...
     @property
-    def state(self):  # -> tuple[Any | array | None, Any | array | None]:
-        ...
+    def state(self) -> tuple[mx.array | None, mx.array | None]: ...
     @state.setter
     def state(self, v):  # -> None:
         ...
     def is_trimmable(self):  # -> Literal[True]:
         ...
-    def trim(self, n):  # -> int:
-        ...
+    def trim(self, n: int) -> int: ...
     def make_mask(self, *args, **kwargs):  # -> array | Literal['causal'] | None:
         ...
 
@@ -132,10 +131,7 @@ class QuantizedKVCache(_BaseCache):
     def update_and_fetch(self, keys, values):  # -> Any:
         ...
     @property
-    def state(
-        self,
-    ):  # -> tuple[Any | tuple[array, array, array] | None, Any | tuple[array, array, array] | None] | Any:
-        ...
+    def state(self) -> tuple[mx.array | None, mx.array | None]: ...
     @state.setter
     def state(self, v):  # -> None:
         ...
@@ -147,8 +143,7 @@ class QuantizedKVCache(_BaseCache):
         ...
     def is_trimmable(self):  # -> Literal[True]:
         ...
-    def trim(self, n):  # -> int:
-        ...
+    def trim(self, n: int) -> int: ...
     def make_mask(self, *args, **kwargs):  # -> array | Literal['causal'] | None:
         ...
 
@@ -160,22 +155,30 @@ class KVCache(_BaseCache):
     @property
     def state(
         self,
-    ) -> tuple[array, array]: ...
+    ) -> tuple[mx.array | None, mx.array | None]: ...
     @state.setter
     def state(self, v) -> None: ...
     def is_trimmable(self):  # -> Literal[True]:
         ...
-    def trim(self, n):  # -> int:
-        ...
+    def trim(self, n: int) -> int: ...
     def to_quantized(
         self, group_size: int = ..., bits: int = ...
     ) -> QuantizedKVCache: ...
-    def make_mask(self, *args, **kwargs):  # -> array | Literal['causal'] | None:
-        ...
+    def make_mask(
+        self, *args: Any, **kwargs: Any
+    ) -> mx.array | Literal["causal"] | None: ...
 
 class RotatingKVCache(_BaseCache):
     step = ...
+    keys: mx.array | None
+    values: mx.array | None
+    keep: int
+    max_size: int
+    _idx: int
     def __init__(self, max_size, keep=...) -> None: ...
+    def _trim(
+        self, trim_size: int, v: mx.array, append: mx.array | None = ...
+    ) -> mx.array: ...
     def update_and_fetch(
         self, keys, values
     ):  # -> tuple[array | Any, array | Any] | tuple[array | Any, array | Any | None]:
@@ -183,8 +186,7 @@ class RotatingKVCache(_BaseCache):
     @property
     def state(
         self,
-    ):  # -> tuple[Any | array, Any | array] | tuple[Any | array | None, Any | array | None]:
-        ...
+    ) -> tuple[mx.array | None, mx.array | None]: ...
     @state.setter
     def state(self, v):  # -> None:
         ...
@@ -196,8 +198,7 @@ class RotatingKVCache(_BaseCache):
         ...
     def is_trimmable(self):  # -> bool:
         ...
-    def trim(self, n):  # -> int:
-        ...
+    def trim(self, n: int) -> int: ...
     def to_quantized(
         self, group_size: int = ..., bits: int = ...
     ) -> QuantizedKVCache: ...
@@ -212,8 +213,7 @@ class ArraysCache(_BaseCache):
         ...
     def __getitem__(self, idx): ...
     @property
-    def state(self):  # -> list[Any | array] | list[array]:
-        ...
+    def state(self) -> tuple[mx.array | None, mx.array | None]: ...
     @state.setter
     def state(self, v):  # -> None:
         ...
@@ -227,8 +227,7 @@ class ArraysCache(_BaseCache):
         In-place extend this cache with the other cache.
         """
 
-    def make_mask(self, N: int):  # -> array | None:
-        ...
+    def make_mask(self, N: int) -> mx.array | None: ...
 
 class MambaCache(ArraysCache):
     def __init__(self, left_padding: Optional[List[int]] = ...) -> None: ...
@@ -239,8 +238,7 @@ class ChunkedKVCache(KVCache):
         ...
     def update_and_fetch(self, keys, values):  # -> tuple[array, array]:
         ...
-    def trim(self, n):  # -> int:
-        ...
+    def trim(self, n: int) -> int: ...
     @property
     def meta_state(self):  # -> tuple[str, ...]:
         ...
@@ -253,10 +251,9 @@ class CacheList(_BaseCache):
     def __getitem__(self, idx): ...
     def is_trimmable(self):  # -> bool:
         ...
-    def trim(self, n): ...
+    def trim(self, n: int) -> int: ...
     @property
-    def state(self):  # -> list[Any]:
-        ...
+    def state(self) -> list[tuple[mx.array | None, mx.array | None]]: ...
     @state.setter
     def state(self, v):  # -> None:
         ...
@@ -271,29 +268,14 @@ class CacheList(_BaseCache):
         """
 
 class BatchKVCache(_BaseCache):
-    step = ...
-    def __init__(self, left_padding: List[int]) -> None:
-        """
-        The BatchKV cache expects inputs to be left-padded.
-
-        E.g. the following prompts:
-
-            [1, 3, 5]
-            [7]
-            [2, 6, 8, 9]
-
-        Should be padded like so:
-
-            [0, 1, 3, 5]
-            [0, 0, 0, 7]
-            [2, 6, 8, 9]
-
-        And ``left_padding`` specifies the amount of padding for each.
-        In this case, ``left_padding = [1, 3, 0]``.
-        """
-
-    def update_and_fetch(self, keys, values):  # -> tuple[array | Any, array | Any]:
-        ...
+    step: int
+    keys: array | None
+    values: array | None
+    offset: array
+    left_padding: array
+    _idx: int
+    def __init__(self, left_padding: List[int]) -> None: ...
+    def update_and_fetch(self, keys: array, values: array) -> tuple[array, array]: ...
     @property
     def state(
         self,
@@ -319,12 +301,21 @@ class BatchKVCache(_BaseCache):
         """
 
 class BatchRotatingKVCache(_BaseCache):
-    step = ...
-    def __init__(self, max_size, left_padding: List[int]) -> None: ...
-    def update_and_fetch(
-        self, keys, values
-    ):  # -> tuple[array | Any, array | Any] | tuple[array | Any, array | Any | None]:
-        ...
+    step: int
+    keys: array | None
+    values: array | None
+    offset: array
+    left_padding: array
+    max_size: int
+    _idx: int
+    _offset: int
+    rotated: bool
+    _lengths: array | None
+    def __init__(self, max_size: int, left_padding: List[int]) -> None: ...
+    def _trim(self, trim_size: int, v: array, append: array | None = ...) -> array: ...
+    def _update_in_place(self, keys: array, values: array) -> tuple[array, array]: ...
+    def _update_concat(self, keys: array, values: array) -> tuple[array, array]: ...
+    def update_and_fetch(self, keys: array, values: array) -> tuple[array, array]: ...
     @property
     def state(
         self,

.mlx_typings/mlx_lm/models/gated_delta.pyi

@@ -0,0 +1,35 @@
+from typing import Optional
+
+import mlx.core as mx
+
+def compute_g(A_log: mx.array, a: mx.array, dt_bias: mx.array) -> mx.array: ...
+def gated_delta_update(
+    q: mx.array,
+    k: mx.array,
+    v: mx.array,
+    a: mx.array,
+    b: mx.array,
+    A_log: mx.array,
+    dt_bias: mx.array,
+    state: Optional[mx.array] = ...,
+    mask: Optional[mx.array] = ...,
+    use_kernel: bool = ...,
+) -> tuple[mx.array, mx.array]: ...
+def gated_delta_ops(
+    q: mx.array,
+    k: mx.array,
+    v: mx.array,
+    g: mx.array,
+    beta: mx.array,
+    state: Optional[mx.array] = ...,
+    mask: Optional[mx.array] = ...,
+) -> tuple[mx.array, mx.array]: ...
+def gated_delta_kernel(
+    q: mx.array,
+    k: mx.array,
+    v: mx.array,
+    g: mx.array,
+    beta: mx.array,
+    state: mx.array,
+    mask: Optional[mx.array] = ...,
+) -> tuple[mx.array, mx.array]: ...

.mlx_typings/mlx_lm/models/glm_moe_dsa.pyi

@@ -0,0 +1,46 @@
+"""Type stubs for mlx_lm.models.glm_moe_dsa"""
+
+from dataclasses import dataclass
+from typing import Any, Dict, Optional
+
+from .base import BaseModelArgs
+from .deepseek_v32 import Model as DSV32Model
+
+@dataclass
+class ModelArgs(BaseModelArgs):
+    model_type: str
+    vocab_size: int
+    hidden_size: int
+    index_head_dim: int
+    index_n_heads: int
+    index_topk: int
+    intermediate_size: int
+    moe_intermediate_size: int
+    num_hidden_layers: int
+    num_attention_heads: int
+    num_key_value_heads: int
+    n_shared_experts: Optional[int]
+    n_routed_experts: Optional[int]
+    routed_scaling_factor: float
+    kv_lora_rank: int
+    q_lora_rank: int
+    qk_rope_head_dim: int
+    v_head_dim: int
+    qk_nope_head_dim: int
+    topk_method: str
+    scoring_func: str
+    norm_topk_prob: bool
+    n_group: int
+    topk_group: int
+    num_experts_per_tok: int
+    moe_layer_freq: int
+    first_k_dense_replace: int
+    max_position_embeddings: int
+    rms_norm_eps: float
+    rope_parameters: Dict[str, Any]
+    attention_bias: bool
+    rope_scaling: Dict[str, Any] | None
+    rope_theta: float | None
+
+class Model(DSV32Model):
+    def __init__(self, config: ModelArgs) -> None: ...

.mlx_typings/mlx_lm/models/nemotron_h.pyi

@@ -0,0 +1,154 @@
+from dataclasses import dataclass
+from typing import Any, List, Optional, Tuple
+
+import mlx.core as mx
+import mlx.nn as nn
+
+from .cache import ArraysCache, KVCache
+from .switch_layers import SwitchMLP
+
+@dataclass
+class ModelArgs:
+    model_type: str
+    vocab_size: int
+    hidden_size: int
+    intermediate_size: int
+    num_hidden_layers: int
+    max_position_embeddings: int
+    num_attention_heads: int
+    num_key_value_heads: int
+    attention_bias: bool
+    mamba_num_heads: int
+    mamba_head_dim: int
+    mamba_proj_bias: bool
+    ssm_state_size: int
+    conv_kernel: int
+    n_groups: int
+    mlp_bias: bool
+    layer_norm_epsilon: float
+    use_bias: bool
+    use_conv_bias: bool
+    hybrid_override_pattern: List[str]
+    head_dim: Optional[int]
+    moe_intermediate_size: Optional[int]
+    moe_shared_expert_intermediate_size: Optional[int]
+    n_group: Optional[int]
+    n_routed_experts: Optional[int]
+    n_shared_experts: Optional[int]
+    topk_group: Optional[int]
+    num_experts_per_tok: Optional[int]
+    norm_topk_prob: Optional[bool]
+    routed_scaling_factor: Optional[float]
+    time_step_limit: Optional[Tuple[float, float]]
+    time_step_min: Optional[float]
+    time_step_max: Optional[float]
+
+    @classmethod
+    def from_dict(cls, params: dict[str, Any]) -> ModelArgs: ...
+    def __post_init__(self) -> None: ...
+
+class NemotronHMamba2Mixer(nn.Module):
+    num_heads: int
+    hidden_size: int
+    ssm_state_size: int
+    conv_kernel_size: int
+    intermediate_size: int
+    n_groups: int
+    head_dim: int
+    conv_dim: int
+    conv1d: nn.Conv1d
+    in_proj: nn.Linear
+    dt_bias: mx.array
+    A_log: mx.array
+    D: mx.array
+    norm: nn.RMSNorm
+    heads_per_group: int
+    out_proj: nn.Linear
+
+    def __init__(self, args: ModelArgs) -> None: ...
+    def __call__(
+        self,
+        hidden_states: mx.array,
+        mask: Optional[mx.array],
+        cache: Optional[ArraysCache] = None,
+    ) -> mx.array: ...
+
+class NemotronHAttention(nn.Module):
+    hidden_size: int
+    num_heads: int
+    head_dim: int
+    num_key_value_heads: int
+    scale: float
+    q_proj: nn.Linear
+    k_proj: nn.Linear
+    v_proj: nn.Linear
+    o_proj: nn.Linear
+
+    def __init__(self, args: ModelArgs) -> None: ...
+    def __call__(
+        self,
+        x: mx.array,
+        mask: Optional[mx.array] = None,
+        cache: Optional[KVCache] = None,
+    ) -> mx.array: ...
+
+class NemotronHMLP(nn.Module):
+    up_proj: nn.Linear
+    down_proj: nn.Linear
+
+    def __init__(
+        self, args: ModelArgs, intermediate_size: Optional[int] = None
+    ) -> None: ...
+    def __call__(self, x: mx.array) -> mx.array: ...
+
+class NemotronHMoE(nn.Module):
+    num_experts_per_tok: int
+    switch_mlp: SwitchMLP
+    shared_experts: NemotronHMLP
+
+    def __init__(self, config: ModelArgs) -> None: ...
+    def __call__(self, x: mx.array) -> mx.array: ...
+
+class NemotronHBlock(nn.Module):
+    block_type: str
+    norm: nn.RMSNorm
+    mixer: NemotronHMamba2Mixer | NemotronHAttention | NemotronHMLP | NemotronHMoE
+
+    def __init__(self, args: ModelArgs, block_type: str) -> None: ...
+    def __call__(
+        self,
+        x: mx.array,
+        mask: Optional[mx.array] = None,
+        cache: Optional[Any] = None,
+    ) -> mx.array: ...
+
+class NemotronHModel(nn.Module):
+    embeddings: nn.Embedding
+    layers: list[NemotronHBlock]
+    norm_f: nn.RMSNorm
+    fa_idx: int
+    ssm_idx: int
+
+    def __init__(self, args: ModelArgs) -> None: ...
+    def __call__(
+        self,
+        inputs: mx.array,
+        cache: Optional[Any] = None,
+    ) -> mx.array: ...
+
+class Model(nn.Module):
+    args: ModelArgs
+    backbone: NemotronHModel
+    lm_head: nn.Linear
+    model_type: str
+
+    def __init__(self, args: ModelArgs) -> None: ...
+    def __call__(
+        self,
+        inputs: mx.array,
+        cache: Optional[Any] = None,
+    ) -> mx.array: ...
+    @property
+    def layers(self) -> list[NemotronHBlock]: ...
+    def make_cache(self) -> list[ArraysCache | KVCache]: ...
+    def sanitize(self, weights: dict[str, Any]) -> dict[str, Any]: ...

.mlx_typings/mlx_lm/models/qwen3_5.pyi

@@ -0,0 +1,153 @@
+from dataclasses import dataclass
+from typing import Any, Optional
+
+import mlx.core as mx
+import mlx.nn as nn
+
+from .cache import ArraysCache, KVCache
+from .qwen3_next import (
+    Qwen3NextAttention as Attention,
+    Qwen3NextMLP as MLP,
+    Qwen3NextRMSNormGated as RMSNormGated,
+    Qwen3NextSparseMoeBlock,
+)
+
+SparseMoeBlock = Qwen3NextSparseMoeBlock
+from .switch_layers import SwitchGLU
+
+@dataclass
+class TextModelArgs:
+    model_type: str
+    hidden_size: int
+    intermediate_size: int
+    num_hidden_layers: int
+    num_attention_heads: int
+    rms_norm_eps: float
+    vocab_size: int
+    num_key_value_heads: int
+    max_position_embeddings: int
+    linear_num_value_heads: int
+    linear_num_key_heads: int
+    linear_key_head_dim: int
+    linear_value_head_dim: int
+    linear_conv_kernel_dim: int
+    tie_word_embeddings: bool
+    attention_bias: bool
+    head_dim: Optional[int]
+    full_attention_interval: int
+    num_experts: int
+    num_experts_per_tok: int
+    decoder_sparse_step: int
+    shared_expert_intermediate_size: int
+    moe_intermediate_size: int
+    norm_topk_prob: bool
+    rope_parameters: Optional[dict[str, Any]]
+    partial_rotary_factor: float
+    rope_theta: float
+    rope_scaling: Optional[dict[str, Any]]
+
+    @classmethod
+    def from_dict(cls, params: dict[str, Any]) -> TextModelArgs: ...
+    def __post_init__(self) -> None: ...
+
+class GatedDeltaNet(nn.Module):
+    hidden_size: int
+    num_v_heads: int
+    num_k_heads: int
+    head_k_dim: int
+    head_v_dim: int
+    key_dim: int
+    value_dim: int
+    conv_kernel_size: int
+    conv_dim: int
+    conv1d: nn.Conv1d
+    in_proj_qkv: nn.Linear
+    in_proj_z: nn.Linear
+    in_proj_b: nn.Linear
+    in_proj_a: nn.Linear
+    dt_bias: mx.array
+    A_log: mx.array
+    norm: RMSNormGated
+    out_proj: nn.Linear
+
+    def __init__(self, config: TextModelArgs) -> None: ...
+    def __call__(
+        self,
+        inputs: mx.array,
+        mask: Optional[mx.array] = None,
+        cache: Optional[Any] = None,
+    ) -> mx.array: ...
+
+class DecoderLayer(nn.Module):
+    is_linear: bool
+    linear_attn: GatedDeltaNet
+    self_attn: Attention
+    input_layernorm: nn.RMSNorm
+    post_attention_layernorm: nn.RMSNorm
+    mlp: MLP | SparseMoeBlock
+
+    def __init__(self, args: TextModelArgs, layer_idx: int) -> None: ...
+    def __call__(
+        self,
+        x: mx.array,
+        mask: Optional[mx.array] = None,
+        cache: Optional[Any] = None,
+    ) -> mx.array: ...
+
+class Qwen3_5TextModel(nn.Module):
+    embed_tokens: nn.Embedding
+    layers: list[DecoderLayer]
+    norm: nn.RMSNorm
+    ssm_idx: int
+    fa_idx: int
+
+    def __init__(self, args: TextModelArgs) -> None: ...
+    def __call__(
+        self,
+        inputs: mx.array,
+        cache: Optional[Any] = None,
+        input_embeddings: Optional[mx.array] = None,
+    ) -> mx.array: ...
+
+class TextModel(nn.Module):
+    args: TextModelArgs
+    model_type: str
+    model: Qwen3_5TextModel
+    lm_head: nn.Linear
+
+    def __init__(self, args: TextModelArgs) -> None: ...
+    def __call__(
+        self,
+        inputs: mx.array,
+        cache: Optional[Any] = None,
+        input_embeddings: Optional[mx.array] = None,
+    ) -> mx.array: ...
+    @property
+    def layers(self) -> list[DecoderLayer]: ...
+    def make_cache(self) -> list[ArraysCache | KVCache]: ...
+    def sanitize(self, weights: dict[str, Any]) -> dict[str, Any]: ...
+
+@dataclass
+class ModelArgs:
+    model_type: str
+    text_config: dict[str, Any]
+
+    @classmethod
+    def from_dict(cls, params: dict[str, Any]) -> ModelArgs: ...
+
+class Model(nn.Module):
+    args: ModelArgs
+    model_type: str
+    language_model: TextModel
+
+    def __init__(self, args: ModelArgs) -> None: ...
+    def __call__(
+        self,
+        inputs: mx.array,
+        cache: Optional[Any] = None,
+        input_embeddings: Optional[mx.array] = None,
+    ) -> mx.array: ...
+    def sanitize(self, weights: dict[str, Any]) -> dict[str, Any]: ...
+    @property
+    def layers(self) -> list[DecoderLayer]: ...
+    def make_cache(self) -> list[ArraysCache | KVCache]: ...

.mlx_typings/mlx_lm/models/qwen3_5_moe.pyi

@@ -0,0 +1,19 @@
+from dataclasses import dataclass
+from typing import Any, Optional
+
+import mlx.core as mx
+import mlx.nn as nn
+
+from .cache import ArraysCache, KVCache
+from .qwen3_5 import DecoderLayer, Model as Qwen3_5Model, TextModel
+
+@dataclass
+class ModelArgs:
+    model_type: str
+    text_config: dict[str, Any]
+
+    @classmethod
+    def from_dict(cls, params: dict[str, Any]) -> ModelArgs: ...
+
+class Model(Qwen3_5Model):
+    def sanitize(self, weights: dict[str, Any]) -> dict[str, Any]: ...

.mlx_typings/mlx_lm/models/qwen3_next.pyi

@@ -5,8 +5,18 @@ from typing import Any, Optional
 import mlx.core as mx
 import mlx.nn as nn
 
+from .cache import ArraysCache, KVCache
 from .switch_layers import SwitchGLU
 
+class Qwen3NextRMSNormGated(nn.Module):
+    eps: float
+    weight: mx.array
+
+    def __init__(self, hidden_size: int, eps: float = ...) -> None: ...
+    def __call__(
+        self, hidden_states: mx.array, gate: mx.array | None = None
+    ) -> mx.array: ...
+
 class Qwen3NextMLP(nn.Module):
     gate_proj: nn.Linear
     down_proj: nn.Linear
@@ -90,6 +100,8 @@ class Qwen3NextModel(nn.Module):
     embed_tokens: nn.Embedding
     layers: list[Qwen3NextDecoderLayer]
     norm: nn.RMSNorm
+    ssm_idx: int
+    fa_idx: int
 
     def __init__(self, args: Any) -> None: ...
     def __call__(
@@ -112,3 +124,4 @@ class Model(nn.Module):
     def sanitize(self, weights: dict[str, Any]) -> dict[str, Any]: ...
     @property
     def layers(self) -> list[Qwen3NextDecoderLayer]: ...
+    def make_cache(self) -> list[ArraysCache | KVCache]: ...

.mlx_typings/mlx_lm/models/rope_utils.pyi

@@ -0,0 +1,51 @@
+from typing import Any, Optional
+
+import mlx.nn as nn
+
+class YarnRoPE(nn.Module):
+    def __init__(
+        self,
+        dims: int,
+        traditional: bool = ...,
+        max_position_embeddings: int = ...,
+        base: float = ...,
+        scaling_factor: float = ...,
+        original_max_position_embeddings: int = ...,
+        beta_fast: float = ...,
+        beta_slow: float = ...,
+        mscale: float = ...,
+        mscale_all_dim: float = ...,
+    ) -> None: ...
+
+class Llama3RoPE(nn.Module):
+    def __init__(
+        self,
+        dims: int,
+        traditional: bool = ...,
+        max_position_embeddings: int = ...,
+        base: float = ...,
+        scaling_factor: float = ...,
+        original_max_position_embeddings: int = ...,
+        low_freq_factor: float = ...,
+        high_freq_factor: float = ...,
+    ) -> None: ...
+
+class SuScaledRoPE(nn.Module):
+    def __init__(
+        self,
+        dims: int,
+        traditional: bool = ...,
+        max_position_embeddings: int = ...,
+        base: float = ...,
+        short_factor: Any = ...,
+        long_factor: Any = ...,
+        original_max_position_embeddings: int = ...,
+    ) -> None: ...
+
+def initialize_rope(
+    dims: int,
+    base: float = ...,
+    traditional: bool = ...,
+    scaling_config: Optional[dict[str, Any]] = ...,
+    max_position_embeddings: Optional[int] = ...,
+) -> nn.Module: ...

.mlx_typings/mlx_lm/models/switch_layers.pyi

@@ -73,6 +73,9 @@ class SwitchGLU(nn.Module):
     def __call__(self, x, indices) -> mx.array: ...
 
 class SwitchMLP(nn.Module):
+    fc1: SwitchLinear
+    fc2: SwitchLinear
+
     def __init__(
         self,
         input_dims: int,

.mlx_typings/mlx_lm/sample_utils.pyi

@@ -48,7 +48,7 @@ def make_logits_processors(
     logit_bias: Optional[Dict[int, float]] = ...,
     repetition_penalty: Optional[float] = ...,
     repetition_context_size: Optional[int] = ...,
-):  # -> list[Any]:
+) -> list[Callable[[mx.array, mx.array], mx.array]]:
     """
     Make logits processors for use with ``generate_step``.
 

.mlx_typings/mlx_lm/tokenizer_utils.pyi

@@ -39,11 +39,11 @@ class StreamingDetokenizer:
     """
 
     __slots__ = ...
-    def reset(self): ...
-    def add_token(self, token): ...
-    def finalize(self): ...
+    def reset(self) -> None: ...
+    def add_token(self, token: int) -> None: ...
+    def finalize(self) -> None: ...
     @property
-    def last_segment(self):
+    def last_segment(self) -> str:
         """Return the last segment of readable text since last time this property was accessed."""
 
 class NaiveStreamingDetokenizer(StreamingDetokenizer):

CONTRIBUTING.md

@@ -11,9 +11,18 @@ To run EXO from source:
   ```bash
   brew install uv
   ```
-- [macmon](https://github.com/vladkens/macmon) (for hardware monitoring on Apple Silicon)
+- [rust](https://github.com/rust-lang/rustup) (to build Rust bindings, nightly for now)
   ```bash
-  brew install macmon
+  curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
+  rustup toolchain install nightly
+  ```
+- [macmon](https://github.com/vladkens/macmon) (for hardware monitoring on Apple Silicon)
+  Use the pinned fork revision used by this repo instead of Homebrew `macmon`.
+  ```bash
+  cargo install --git https://github.com/swiftraccoon/macmon \
+    --rev 9154d234f763fbeffdcb4135d0bbbaf80609699b \
+    macmon \
+    --force
   ```
 
 ```bash
@@ -39,6 +48,119 @@ Write pure functions where possible. When adding new code, prefer Rust unless th
 
 Run `nix fmt` to auto-format your code before submitting.
 
+## Model Cards
+
+EXO uses TOML-based model cards to define model metadata and capabilities. Model cards are stored in:
+- `resources/inference_model_cards/` for text generation models
+- `resources/image_model_cards/` for image generation models
+- `~/.exo/custom_model_cards/` for user-added custom models
+
+### Adding a Model Card
+
+To add a new model, create a TOML file with the following structure:
+
+```toml
+model_id = "mlx-community/Llama-3.2-1B-Instruct-4bit"
+n_layers = 16
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "llama"
+quantization = "4bit"
+base_model = "Llama 3.2 1B"
+capabilities = ["text"]
+
+[storage_size]
+in_bytes = 729808896
+```
+
+### Required Fields
+
+- `model_id`: Hugging Face model identifier
+- `n_layers`: Number of transformer layers
+- `hidden_size`: Hidden dimension size
+- `supports_tensor`: Whether the model supports tensor parallelism
+- `tasks`: List of supported tasks (`TextGeneration`, `TextToImage`, `ImageToImage`)
+- `family`: Model family (e.g., "llama", "deepseek", "qwen")
+- `quantization`: Quantization level (e.g., "4bit", "8bit", "bf16")
+- `base_model`: Human-readable base model name
+- `capabilities`: List of capabilities (e.g., `["text"]`, `["text", "thinking"]`)
+
+### Optional Fields
+
+- `components`: For multi-component models (like image models with separate text encoders and transformers)
+- `uses_cfg`: Whether the model uses classifier-free guidance (for image models)
+- `trust_remote_code`: Whether to allow remote code execution (defaults to `false` for security)
+
+### Capabilities
+
+The `capabilities` field defines what the model can do:
+- `text`: Standard text generation
+- `thinking`: Model supports chain-of-thought reasoning
+- `thinking_toggle`: Thinking can be enabled/disabled via `enable_thinking` parameter
+- `image_edit`: Model supports image-to-image editing (FLUX.1-Kontext)
+
+### Security Note
+
+By default, `trust_remote_code` is set to `false` for security. Only enable it if the model explicitly requires remote code execution from the Hugging Face hub.
+
+## API Adapters
+
+EXO supports multiple API formats through an adapter pattern. Adapters convert API-specific request formats to the internal `TextGenerationTaskParams` format and convert internal token chunks back to API-specific responses.
+
+### Adapter Architecture
+
+All adapters live in `src/exo/master/adapters/` and follow the same pattern:
+
+1. Convert API-specific requests to `TextGenerationTaskParams`
+2. Handle both streaming and non-streaming response generation
+3. Convert internal `TokenChunk` objects to API-specific formats
+4. Manage error handling and edge cases
+
+### Existing Adapters
+
+- `chat_completions.py`: OpenAI Chat Completions API
+- `claude.py`: Anthropic Claude Messages API
+- `responses.py`: OpenAI Responses API
+- `ollama.py`: Ollama API (for OpenWebUI compatibility)
+
+### Adding a New API Adapter
+
+To add support for a new API format:
+
+1. Create a new adapter file in `src/exo/master/adapters/`
+2. Implement a request conversion function:
+   ```python
+   def your_api_request_to_text_generation(
+       request: YourAPIRequest,
+   ) -> TextGenerationTaskParams:
+       # Convert API request to internal format
+       pass
+   ```
+3. Implement streaming response generation:
+   ```python
+   async def generate_your_api_stream(
+       command_id: CommandId,
+       chunk_stream: AsyncGenerator[TokenChunk | ErrorChunk | ToolCallChunk, None],
+   ) -> AsyncGenerator[str, None]:
+       # Convert internal chunks to API-specific streaming format
+       pass
+   ```
+4. Implement non-streaming response collection:
+   ```python
+   async def collect_your_api_response(
+       command_id: CommandId,
+       chunk_stream: AsyncGenerator[TokenChunk | ErrorChunk | ToolCallChunk, None],
+   ) -> AsyncGenerator[str]:
+       # Collect all chunks and return single response
+       pass
+   ```
+5. Register the adapter endpoints in `src/exo/master/api.py`
+
+The adapter pattern keeps API-specific logic isolated from core inference systems. Internal systems (worker, runner, event sourcing) only see `TextGenerationTaskParams` and `TokenChunk` objects - no API-specific types cross the adapter boundary.
+
+For detailed API documentation, see [docs/api.md](docs/api.md).
+
 ## Testing
 
 EXO relies heavily on manual testing at this point in the project, but this is evolving. Before submitting a change, test both before and after to demonstrate how your change improves behavior. Do the best you can with the hardware you have available - if you need help testing, ask and we'll do our best to assist. Add automated tests where possible - we're actively working to substantially improve our automated testing story.

Cargo.lock

@@ -141,12 +141,6 @@ version = "0.3.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "76a2e8124351fda1ef8aaaa3bbd7ebbcb486bbcd4225aca0aa0d84bb2db8fecb"
 
-[[package]]
-name = "arrayvec"
-version = "0.7.6"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7c02d123df017efcdfbd739ef81735b36c5ba83ec3c59c80a9d7ecc718f92e50"
-
 [[package]]
 name = "asn1-rs"
 version = "0.7.1"
@@ -222,6 +216,28 @@ dependencies = [
  "windows-sys 0.61.2",
 ]
 
+[[package]]
+name = "async-stream"
+version = "0.3.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0b5a71a6f37880a80d1d7f19efd781e4b5de42c88f0722cc13bcb6cc2cfe8476"
+dependencies = [
+ "async-stream-impl",
+ "futures-core",
+ "pin-project-lite",
+]
+
+[[package]]
+name = "async-stream-impl"
+version = "0.3.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c7c24de15d275a1ecfd47a380fb4d5ec9bfe0933f309ed5e705b775596a3574d"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.111",
+]
+
 [[package]]
 name = "async-trait"
 version = "0.1.89"
@@ -304,19 +320,6 @@ version = "1.8.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "55248b47b0caf0546f7988906588779981c43bb1bc9d0c44087278f80cdb44ba"
 
-[[package]]
-name = "bigdecimal"
-version = "0.4.9"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "560f42649de9fa436b73517378a147ec21f6c997a546581df4b4b31677828934"
-dependencies = [
- "autocfg",
- "libm",
- "num-bigint",
- "num-integer",
- "num-traits",
-]
-
 [[package]]
 name = "bimap"
 version = "0.6.3"
@@ -516,15 +519,6 @@ version = "0.4.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "2f421161cb492475f1661ddc9815a745a1c894592070661180fdec3d4872e9c3"
 
-[[package]]
-name = "convert_case"
-version = "0.10.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "633458d4ef8c78b72454de2d54fd6ab2e60f9e02be22f3c6104cdc8a4e0fceb9"
-dependencies = [
- "unicode-segmentation",
-]
-
 [[package]]
 name = "core-foundation"
 version = "0.9.4"
@@ -746,29 +740,6 @@ dependencies = [
  "powerfmt",
 ]
 
-[[package]]
-name = "derive_more"
-version = "2.1.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "10b768e943bed7bf2cab53df09f4bc34bfd217cdb57d971e769874c9a6710618"
-dependencies = [
- "derive_more-impl",
-]
-
-[[package]]
-name = "derive_more-impl"
-version = "2.1.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6d286bfdaf75e988b4a78e013ecd79c581e06399ab53fbacd2d916c2f904f30b"
-dependencies = [
- "convert_case",
- "proc-macro2",
- "quote",
- "rustc_version",
- "syn 2.0.111",
- "unicode-xid",
-]
-
 [[package]]
 name = "digest"
 version = "0.10.7"
@@ -939,22 +910,17 @@ name = "exo_pyo3_bindings"
 version = "0.0.1"
 dependencies = [
  "delegate",
- "derive_more",
  "env_logger",
  "extend",
- "futures",
- "impl-trait-for-tuples",
+ "futures-lite",
  "libp2p",
  "log",
  "networking",
- "once_cell",
  "pin-project",
  "pyo3",
  "pyo3-async-runtimes",
  "pyo3-log",
  "pyo3-stub-gen",
- "thiserror 2.0.17",
- "thread_local",
  "tokio",
  "util",
 ]
@@ -970,6 +936,12 @@ dependencies = [
  "syn 2.0.111",
 ]
 
+[[package]]
+name = "fastrand"
+version = "2.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "37909eebbb50d72f9059c3b6d82c0463f2ff062c9e95845c43a6c9c0355411be"
+
 [[package]]
 name = "ff"
 version = "0.13.1"
@@ -1078,7 +1050,10 @@ version = "2.6.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f78e10609fe0e0b3f4157ffab1876319b5b0db102a2c60dc4626306dc46b44ad"
 dependencies = [
+ "fastrand",
  "futures-core",
+ "futures-io",
+ "parking",
  "pin-project-lite",
 ]
 
@@ -1640,17 +1615,6 @@ dependencies = [
  "xmltree",
 ]
 
-[[package]]
-name = "impl-trait-for-tuples"
-version = "0.2.3"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "a0eb5a3343abf848c0984fe4604b2b105da9539376e24fc0a3b0007411ae4fd9"
-dependencies = [
- "proc-macro2",
- "quote",
- "syn 2.0.111",
-]
-
 [[package]]
 name = "indexmap"
 version = "2.12.1"
@@ -1829,12 +1793,6 @@ version = "0.2.178"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "37c93d8daa9d8a012fd8ab92f088405fb202ea0b6ab73ee2482ae66af4f42091"
 
-[[package]]
-name = "libm"
-version = "0.2.15"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f9fbbcab51052fe104eb5e5d351cf728d30a5be1fe14d9be8a3b097481fb97de"
-
 [[package]]
 name = "libp2p"
 version = "0.56.0"
@@ -2823,17 +2781,16 @@ dependencies = [
 name = "networking"
 version = "0.0.1"
 dependencies = [
+ "async-stream",
  "delegate",
- "derive_more",
  "either",
  "extend",
- "futures",
+ "futures-lite",
  "futures-timer",
- "impl-trait-for-tuples",
  "keccak-const",
  "libp2p",
  "log",
- "thiserror 2.0.17",
+ "pin-project",
  "tokio",
  "tracing-subscriber",
  "util",
@@ -2918,17 +2875,6 @@ dependencies = [
  "num-traits",
 ]
 
-[[package]]
-name = "num-rational"
-version = "0.4.2"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f83d14da390562dca69fc84082e73e548e1ad308d24accdedd2720017cb37824"
-dependencies = [
- "num-bigint",
- "num-integer",
- "num-traits",
-]
-
 [[package]]
 name = "num-traits"
 version = "0.2.19"
@@ -3279,28 +3225,14 @@ version = "0.27.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "ab53c047fcd1a1d2a8820fe84f05d6be69e9526be40cb03b73f86b6b03e6d87d"
 dependencies = [
- "bigdecimal",
- "either",
- "hashbrown 0.16.1",
- "indexmap",
  "indoc",
- "inventory",
  "libc",
- "lock_api",
  "memoffset",
- "num-bigint",
- "num-complex",
- "num-rational",
- "num-traits",
  "once_cell",
- "ordered-float",
- "parking_lot",
  "portable-atomic",
  "pyo3-build-config",
  "pyo3-ffi",
  "pyo3-macros",
- "rust_decimal",
- "smallvec",
  "unindent",
 ]
 
@@ -3741,16 +3673,6 @@ dependencies = [
  "tokio",
 ]
 
-[[package]]
-name = "rust_decimal"
-version = "1.39.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "35affe401787a9bd846712274d97654355d21b2a2c092a3139aabe31e9022282"
-dependencies = [
- "arrayvec",
- "num-traits",
-]
-
 [[package]]
 name = "rustc-hash"
 version = "1.1.0"
@@ -4615,24 +4537,12 @@ version = "1.0.22"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9312f7c4f6ff9069b165498234ce8be658059c6728633667c526e27dc2cf1df5"
 
-[[package]]
-name = "unicode-segmentation"
-version = "1.12.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f6ccf251212114b54433ec949fd6a7841275f9ada20dddd2f29e9ceea4501493"
-
 [[package]]
 name = "unicode-width"
 version = "0.2.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b4ac048d71ede7ee76d585517add45da530660ef4390e49b098733c6e897f254"
 
-[[package]]
-name = "unicode-xid"
-version = "0.2.6"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ebc1c04c71510c7f702b52b7c350734c9ff1295c464a03335b00bb84fc54f853"
-
 [[package]]
 name = "unicode_names2"
 version = "1.3.0"

Cargo.toml

@@ -1,10 +1,6 @@
 [workspace]
 resolver = "3"
-members = [
-    "rust/networking",
-    "rust/exo_pyo3_bindings",
-    "rust/util",
-]
+members = ["rust/networking", "rust/exo_pyo3_bindings", "rust/util"]
 
 [workspace.package]
 version = "0.0.1"
@@ -26,49 +22,21 @@ opt-level = 3
 networking = { path = "rust/networking" }
 util = { path = "rust/util" }
 
-# Proc-macro authoring tools
-syn = "2.0"
-quote = "1.0"
-proc-macro2 = "1.0"
-darling = "0.20"
-
 # Macro dependecies
 extend = "1.2"
 delegate = "0.13"
-impl-trait-for-tuples = "0.2"
-clap = "4.5"
-derive_more = { version = "2.0.1", features = ["display"] }
-pin-project = "1"
 
 # Utility dependencies
-itertools = "0.14"
-thiserror = "2"
-internment = "0.8"
-recursion = "0.5"
-regex = "1.11"
-once_cell = "1.21"
-thread_local = "1.1"
-bon = "3.4"
-generativity = "1.1"
-anyhow = "1.0"
 keccak-const = "0.2"
 
-# Functional generics/lenses frameworks
-frunk_core = "0.4"
-frunk = "0.4"
-frunk_utils = "0.2"
-frunk-enum-core = "0.3"
-
 # Async dependencies
+async-stream = "0.3"
 tokio = "1.46"
-futures = "0.3"
-futures-util = "0.3"
+futures-lite = "2.6.1"
 futures-timer = "3.0"
 
 # Data structures
 either = "1.15"
-ordered-float = "5.0"
-ahash = "0.8"
 
 # Tracing/logging
 log = "0.4"

README.md

@@ -26,6 +26,8 @@ exo connects all your devices into an AI cluster. Not only does exo enable runni
 - **Topology-Aware Auto Parallel**: exo figures out the best way to split your model across all available devices based on a realtime view of your device topology. It takes into account device resources and network latency/bandwidth between each link.
 - **Tensor Parallelism**: exo supports sharding models, for up to 1.8x speedup on 2 devices and 3.2x speedup on 4 devices.
 - **MLX Support**: exo uses [MLX](https://github.com/ml-explore/mlx) as an inference backend and [MLX distributed](https://ml-explore.github.io/mlx/build/html/usage/distributed.html) for distributed communication.
+- **Multiple API Compatibility**: Compatible with OpenAI Chat Completions API, Claude Messages API, OpenAI Responses API, and Ollama API - use your existing tools and clients.
+- **Custom Model Support**: Load custom models from HuggingFace hub to expand the range of available models.
 
 ## Dashboard
 
@@ -72,18 +74,31 @@ There are two ways to run exo:
 
 ### Run from Source (macOS)
 
+If you have [Nix](https://nixos.org/) installed, you can skip most of the steps below and run exo directly:
+
+```bash
+nix run .#exo
+```
+
+**Note:** To accept the Cachix binary cache (and avoid the Xcode Metal ToolChain), add to `/etc/nix/nix.conf`:
+```
+trusted-users = root    (or your username)
+experimental-features = nix-command flakes
+```
+Then restart the Nix daemon: `sudo launchctl kickstart -k system/org.nixos.nix-daemon`
+
 **Prerequisites:**
+- [Xcode](https://developer.apple.com/xcode/) (provides the Metal ToolChain required for MLX compilation)
 - [brew](https://github.com/Homebrew/brew) (for simple package management on macOS)
-  
+
   ```bash
   /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
   ```
 - [uv](https://github.com/astral-sh/uv) (for Python dependency management)
-- [macmon](https://github.com/vladkens/macmon) (for hardware monitoring on Apple Silicon)
 - [node](https://github.com/nodejs/node) (for building the dashboard)
-  
+
   ```bash
-  brew install uv macmon node
+  brew install uv node
   ```
 - [rust](https://github.com/rust-lang/rustup) (to build Rust bindings, nightly for now)
 
@@ -91,6 +106,17 @@ There are two ways to run exo:
   curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
   rustup toolchain install nightly
   ```
+- [macmon](https://github.com/vladkens/macmon) (for hardware monitoring on Apple Silicon)
+
+  Install the pinned fork revision used by this repo instead of Homebrew `macmon`.
+  Homebrew `macmon 0.6.1` still crashes on Apple M5.
+
+  ```bash
+  cargo install --git https://github.com/swiftraccoon/macmon \
+    --rev 9154d234f763fbeffdcb4135d0bbbaf80609699b \
+    macmon \
+    --force
+  ```
 
 Clone the repo, build the dashboard, and run exo:
 
@@ -182,6 +208,8 @@ exo follows the [XDG Base Directory Specification](https://specifications.freede
 - **Configuration files**: `~/.config/exo/` (or `$XDG_CONFIG_HOME/exo/`)
 - **Data files**: `~/.local/share/exo/` (or `$XDG_DATA_HOME/exo/`)
 - **Cache files**: `~/.cache/exo/` (or `$XDG_CACHE_HOME/exo/`)
+- **Log files**: `~/.cache/exo/exo_log/` (with automatic log rotation)
+- **Custom model cards**: `~/.local/share/exo/custom_model_cards/`
 
 You can override these locations by setting the corresponding XDG environment variables.
 
@@ -261,8 +289,51 @@ After that, RDMA will be enabled in macOS and exo will take care of the rest.
 
 ---
 
+## Environment Variables
+
+exo supports several environment variables for configuration:
+
+| Variable | Description | Default |
+|----------|-------------|---------|
+| `EXO_DEFAULT_MODELS_DIR` | Default directory for model downloads and caches. Always first in the writable dirs list. | `~/.local/share/exo/models` (Linux) or `~/.exo/models` (macOS) |
+| `EXO_MODELS_DIRS` | Colon-separated additional writable directories for model downloads. Checked in order after the default; first with enough free space is used. | None |
+| `EXO_MODELS_READ_ONLY_DIRS` | Colon-separated read-only directories to search for pre-downloaded models (e.g., NFS mounts, shared storage). Models here cannot be deleted. | None |
+| `EXO_OFFLINE` | Run without internet connection (uses only local models) | `false` |
+| `EXO_ENABLE_IMAGE_MODELS` | Enable image model support | `false` |
+| `EXO_LIBP2P_NAMESPACE` | Custom namespace for cluster isolation | None |
+| `EXO_FAST_SYNCH` | Control MLX_METAL_FAST_SYNCH behavior (for JACCL backend) | Auto |
+| `EXO_TRACING_ENABLED` | Enable distributed tracing for performance analysis | `false` |
+
+**Example usage:**
+
+```bash
+# Use pre-downloaded models from NFS mount (read-only)
+EXO_MODELS_READ_ONLY_DIRS=/mnt/nfs/models:/opt/ai-models uv run exo
+
+# Download models to an external SSD (falls back to default dir if full)
+EXO_MODELS_DIRS=/Volumes/ExternalSSD/exo-models uv run exo
+
+# Run in offline mode
+EXO_OFFLINE=true uv run exo
+
+# Enable image models
+EXO_ENABLE_IMAGE_MODELS=true uv run exo
+
+# Use custom namespace for cluster isolation
+EXO_LIBP2P_NAMESPACE=my-dev-cluster uv run exo
+```
+
+---
+
 ### Using the API
 
+exo provides multiple API-compatible interfaces for maximum compatibility with existing tools:
+
+- **OpenAI Chat Completions API** - Compatible with OpenAI clients
+- **Claude Messages API** - Compatible with Anthropic's Claude format
+- **OpenAI Responses API** - Compatible with OpenAI's Responses format
+- **Ollama API** - Compatible with Ollama and tools like OpenWebUI
+
 If you prefer to interact with exo via the API, here is an example creating an instance of a small model (`mlx-community/Llama-3.2-1B-Instruct-4bit`), sending a chat completions request and deleting the instance.
 
 ---
@@ -352,14 +423,85 @@ When you're done, delete the instance by its ID (find it via `/state` or `/insta
 curl -X DELETE http://localhost:52415/instance/YOUR_INSTANCE_ID
 ```
 
+### Claude Messages API Compatibility
+
+Use the Claude Messages API format with the `/v1/messages` endpoint:
+
+```bash
+curl -N -X POST http://localhost:52415/v1/messages \
+  -H 'Content-Type: application/json' \
+  -d '{
+    "model": "mlx-community/Llama-3.2-1B-Instruct-4bit",
+    "messages": [
+      {"role": "user", "content": "Hello"}
+    ],
+    "max_tokens": 1024,
+    "stream": true
+  }'
+```
+
+### OpenAI Responses API Compatibility
+
+Use the OpenAI Responses API format with the `/v1/responses` endpoint:
+
+```bash
+curl -N -X POST http://localhost:52415/v1/responses \
+  -H 'Content-Type: application/json' \
+  -d '{
+    "model": "mlx-community/Llama-3.2-1B-Instruct-4bit",
+    "messages": [
+      {"role": "user", "content": "Hello"}
+    ],
+    "stream": true
+  }'
+```
+
+### Ollama API Compatibility
+
+exo supports Ollama API endpoints for compatibility with tools like OpenWebUI:
+
+```bash
+# Ollama chat
+curl -X POST http://localhost:52415/ollama/api/chat \
+  -H 'Content-Type: application/json' \
+  -d '{
+    "model": "mlx-community/Llama-3.2-1B-Instruct-4bit",
+    "messages": [
+      {"role": "user", "content": "Hello"}
+    ],
+    "stream": false
+  }'
+
+# List models (Ollama format)
+curl http://localhost:52415/ollama/api/tags
+```
+
+### Custom Model Loading from HuggingFace
+
+You can add custom models from the HuggingFace hub:
+
+```bash
+curl -X POST http://localhost:52415/models/add \
+  -H 'Content-Type: application/json' \
+  -d '{
+    "model_id": "mlx-community/my-custom-model"
+  }'
+```
+
+**Security Note:**
+
+Custom models requiring `trust_remote_code` in their configuration must be explicitly enabled (default is false) for security. Only enable this if you trust the model's remote code execution. Models are fetched from HuggingFace and stored locally as custom model cards.
+
 **Other useful API endpoints*:**
 
 - List all models: `curl http://localhost:52415/models`
+- List downloaded models only: `curl http://localhost:52415/models?status=downloaded`
+- Search HuggingFace: `curl "http://localhost:52415/models/search?query=llama&limit=10"`
 - Inspect instance IDs and deployment state: `curl http://localhost:52415/state`
 
 For further details, see:
 
-- API basic documentation in [docs/api.md](docs/api.md).
+- API documentation in [docs/api.md](docs/api.md).
 - API types and endpoints in [src/exo/master/api.py](src/exo/master/api.py).
 
 ---
@@ -418,4 +560,4 @@ On macOS, exo uses the GPU. On Linux, exo currently runs on CPU. We are working
 
 ## Contributing
 
-See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines on how to contribute to exo.
+See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines on how to contribute to exo.

app/EXO/EXO/ContentView.swift

@@ -15,14 +15,23 @@ struct ContentView: View {
     @EnvironmentObject private var localNetworkChecker: LocalNetworkChecker
     @EnvironmentObject private var updater: SparkleUpdater
     @EnvironmentObject private var thunderboltBridgeService: ThunderboltBridgeService
+    @EnvironmentObject private var settingsWindowController: SettingsWindowController
     @State private var focusedNode: NodeViewModel?
     @State private var deletingInstanceIDs: Set<String> = []
     @State private var showAllNodes = false
     @State private var showAllInstances = false
+    @State private var baseURLCopied = false
     @State private var showAdvanced = false
     @State private var showDebugInfo = false
-    @State private var bugReportInFlight = false
-    @State private var bugReportMessage: String?
+    private enum BugReportPhase: Equatable {
+        case idle
+        case prompting
+        case sending(String)
+        case success(String)
+        case failure(String)
+    }
+    @State private var bugReportPhase: BugReportPhase = .idle
+    @State private var bugReportUserDescription: String = ""
     @State private var uninstallInProgress = false
     @State private var pendingNamespace: String = ""
     @State private var pendingHFToken: String = ""
@@ -258,139 +267,79 @@ struct ContentView: View {
         VStack(alignment: .leading, spacing: 0) {
             if controller.status != .stopped {
                 dashboardButton
+                baseURLRow
                 Divider()
                     .padding(.vertical, 8)
             } else {
                 Divider()
                     .padding(.vertical, 4)
             }
-            advancedSection
-                .padding(.bottom, 8)
-            controlButton(title: "Quit", tint: .secondary) {
+            HoverButton(
+                title: "Settings",
+                tint: .primary,
+                trailingSystemImage: "gear"
+            ) {
+                settingsWindowController.open(
+                    controller: controller,
+                    updater: updater,
+                    networkStatusService: networkStatusService,
+                    thunderboltBridgeService: thunderboltBridgeService,
+                    stateService: stateService
+                )
+            }
+            HoverButton(
+                title: "Check for Updates",
+                tint: .primary,
+                trailingSystemImage: "arrow.triangle.2.circlepath"
+            ) {
+                updater.checkForUpdates()
+            }
+            .padding(.bottom, 8)
+            HoverButton(title: "Quit", tint: .secondary) {
                 controller.stop()
                 NSApplication.shared.terminate(nil)
             }
         }
     }
 
-    private var advancedSection: some View {
-        VStack(alignment: .leading, spacing: 6) {
-            HStack {
-                Text("Advanced")
-                    .font(.caption)
-                    .foregroundColor(.secondary)
-                Spacer()
-                collapseButton(isExpanded: $showAdvanced)
-            }
-            .animation(nil, value: showAdvanced)
-            if showAdvanced {
-                VStack(alignment: .leading, spacing: 8) {
-                    VStack(alignment: .leading, spacing: 4) {
-                        Text("Cluster Namespace")
-                            .font(.caption2)
-                            .foregroundColor(.secondary)
-                        HStack {
-                            TextField("optional", text: $pendingNamespace)
-                                .textFieldStyle(.roundedBorder)
-                                .font(.caption2)
-                                .onAppear {
-                                    pendingNamespace = controller.customNamespace
-                                }
-                            Button("Save & Restart") {
-                                controller.customNamespace = pendingNamespace
-                                if controller.status == .running || controller.status == .starting {
-                                    controller.restart()
-                                }
-                            }
-                            .font(.caption2)
-                            .disabled(pendingNamespace == controller.customNamespace)
-                        }
-                    }
-                    VStack(alignment: .leading, spacing: 4) {
-                        Text("HuggingFace Token")
-                            .font(.caption2)
-                            .foregroundColor(.secondary)
-                        HStack {
-                            SecureField("optional", text: $pendingHFToken)
-                                .textFieldStyle(.roundedBorder)
-                                .font(.caption2)
-                                .onAppear {
-                                    pendingHFToken = controller.hfToken
-                                }
-                            Button("Save & Restart") {
-                                controller.hfToken = pendingHFToken
-                                if controller.status == .running || controller.status == .starting {
-                                    controller.restart()
-                                }
-                            }
-                            .font(.caption2)
-                            .disabled(pendingHFToken == controller.hfToken)
-                        }
-                    }
-                    Divider()
-                    HStack {
-                        Toggle(
-                            "Enable Image Models (experimental)", isOn: $pendingEnableImageModels
-                        )
-                        .toggleStyle(.switch)
-                        .font(.caption2)
-                        .onAppear {
-                            pendingEnableImageModels = controller.enableImageModels
-                        }
-
-                        Spacer()
-
-                        Button("Save & Restart") {
-                            controller.enableImageModels = pendingEnableImageModels
-                            if controller.status == .running || controller.status == .starting {
-                                controller.restart()
-                            }
-                        }
-                        .font(.caption2)
-                        .disabled(pendingEnableImageModels == controller.enableImageModels)
-                    }
-                    HoverButton(title: "Check for Updates", small: true) {
-                        updater.checkForUpdates()
-                    }
-                    debugSection
-                    HoverButton(title: "Uninstall", tint: .red, small: true) {
-                        showUninstallConfirmationAlert()
-                    }
-                    .disabled(uninstallInProgress)
-                }
-                .transition(.opacity)
-            }
-        }
-        .animation(.easeInOut(duration: 0.25), value: showAdvanced)
-    }
-
-    private func controlButton(title: String, tint: Color = .primary, action: @escaping () -> Void)
-        -> some View
-    {
-        HoverButton(title: title, tint: tint, trailingSystemImage: nil, action: action)
-    }
-
     private var dashboardButton: some View {
-        Button {
+        HoverButton(
+            title: "Web Dashboard",
+            tint: .primary,
+            trailingSystemImage: "arrow.up.right"
+        ) {
             guard let url = URL(string: "http://localhost:52415/") else { return }
             NSWorkspace.shared.open(url)
-        } label: {
-            HStack {
-                Image(systemName: "arrow.up.right.square")
-                    .imageScale(.small)
-                Text("Dashboard")
-                    .fontWeight(.medium)
-                Spacer()
-            }
-            .padding(.vertical, 8)
-            .padding(.horizontal, 10)
-            .background(
-                RoundedRectangle(cornerRadius: 8, style: .continuous)
-                    .fill(Color(red: 1.0, green: 0.87, blue: 0.0).opacity(0.2))
-            )
         }
-        .buttonStyle(.plain)
-        .padding(.bottom, 4)
+    }
+
+    private var baseURLRow: some View {
+        HStack(spacing: 6) {
+            Image(systemName: "link")
+                .imageScale(.small)
+                .foregroundColor(.secondary)
+            Text("localhost:52415/v1")
+                .font(.system(.caption, design: .monospaced))
+                .foregroundColor(.primary)
+            Spacer()
+            Button {
+                NSPasteboard.general.clearContents()
+                NSPasteboard.general.setString("http://localhost:52415/v1", forType: .string)
+                baseURLCopied = true
+                DispatchQueue.main.asyncAfter(deadline: .now() + 2) {
+                    baseURLCopied = false
+                }
+            } label: {
+                Image(systemName: baseURLCopied ? "checkmark" : "doc.on.doc")
+                    .imageScale(.small)
+                    .foregroundColor(baseURLCopied ? .green : .secondary)
+                    .contentTransition(.symbolEffect(.replace))
+            }
+            .buttonStyle(.plain)
+            .help("Copy API base URL")
+        }
+        .padding(.vertical, 4)
+        .padding(.horizontal, 8)
     }
 
     private func collapseButton(isExpanded: Binding<Bool>) -> some View {
@@ -611,39 +560,115 @@ struct ContentView: View {
     }
 
     private var sendBugReportButton: some View {
-        VStack(alignment: .leading, spacing: 4) {
-            Button {
-                Task {
-                    await sendBugReport()
-                }
-            } label: {
-                HStack {
-                    if bugReportInFlight {
-                        ProgressView()
-                            .scaleEffect(0.6)
+        VStack(alignment: .leading, spacing: 6) {
+            switch bugReportPhase {
+            case .idle:
+                Button {
+                    bugReportPhase = .prompting
+                    bugReportUserDescription = ""
+                } label: {
+                    HStack {
+                        Text("Send Bug Report")
+                            .font(.caption)
+                            .fontWeight(.semibold)
+                        Spacer()
                     }
-                    Text("Send Bug Report")
-                        .font(.caption)
-                        .fontWeight(.semibold)
-                    Spacer()
+                    .padding(.vertical, 6)
+                    .padding(.horizontal, 8)
+                    .background(
+                        RoundedRectangle(cornerRadius: 6)
+                            .fill(Color.accentColor.opacity(0.12))
+                    )
                 }
-                .padding(.vertical, 6)
-                .padding(.horizontal, 8)
+                .buttonStyle(.plain)
+
+            case .prompting:
+                VStack(alignment: .leading, spacing: 6) {
+                    Text("What's the issue? (optional)")
+                        .font(.caption2)
+                        .foregroundColor(.secondary)
+                    TextEditor(text: $bugReportUserDescription)
+                        .font(.caption2)
+                        .frame(height: 60)
+                        .overlay(
+                            RoundedRectangle(cornerRadius: 4)
+                                .stroke(Color.secondary.opacity(0.3), lineWidth: 1)
+                        )
+                    HStack(spacing: 8) {
+                        Button("Send") {
+                            Task {
+                                await sendBugReport()
+                            }
+                        }
+                        .font(.caption2)
+                        .buttonStyle(.borderedProminent)
+                        .controlSize(.small)
+                        Button("Cancel") {
+                            bugReportPhase = .idle
+                        }
+                        .font(.caption2)
+                        .buttonStyle(.bordered)
+                        .controlSize(.small)
+                    }
+                }
+                .padding(8)
                 .background(
                     RoundedRectangle(cornerRadius: 6)
-                        .fill(Color.accentColor.opacity(0.12))
+                        .fill(Color.accentColor.opacity(0.06))
                 )
-            }
-            .buttonStyle(.plain)
-            .disabled(bugReportInFlight)
 
-            if let message = bugReportMessage {
-                Text(message)
+            case .sending(let message):
+                HStack(spacing: 6) {
+                    ProgressView()
+                        .scaleEffect(0.6)
+                    Text(message)
+                        .font(.caption2)
+                        .foregroundColor(.secondary)
+                }
+
+            case .success(let message):
+                VStack(alignment: .leading, spacing: 6) {
+                    Text(message)
+                        .font(.caption2)
+                        .foregroundColor(.secondary)
+                        .fixedSize(horizontal: false, vertical: true)
+                    Button {
+                        openGitHubIssue()
+                    } label: {
+                        HStack(spacing: 4) {
+                            Image(systemName: "arrow.up.right.square")
+                                .imageScale(.small)
+                            Text("Create GitHub Issue")
+                                .font(.caption2)
+                        }
+                    }
+                    .buttonStyle(.bordered)
+                    .controlSize(.small)
+                    Button("Done") {
+                        bugReportPhase = .idle
+                        bugReportUserDescription = ""
+                    }
                     .font(.caption2)
+                    .buttonStyle(.plain)
                     .foregroundColor(.secondary)
-                    .fixedSize(horizontal: false, vertical: true)
+                }
+
+            case .failure(let message):
+                VStack(alignment: .leading, spacing: 4) {
+                    Text(message)
+                        .font(.caption2)
+                        .foregroundColor(.red)
+                        .fixedSize(horizontal: false, vertical: true)
+                    Button("Dismiss") {
+                        bugReportPhase = .idle
+                    }
+                    .font(.caption2)
+                    .buttonStyle(.plain)
+                    .foregroundColor(.secondary)
+                }
             }
         }
+        .animation(.easeInOut(duration: 0.2), value: bugReportPhase)
     }
 
     private var processToggleBinding: Binding<Bool> {
@@ -687,16 +712,58 @@ struct ContentView: View {
     }
 
     private func sendBugReport() async {
-        bugReportInFlight = true
-        bugReportMessage = "Collecting logs..."
+        bugReportPhase = .sending("Collecting logs...")
         let service = BugReportService()
+        let description = bugReportUserDescription.trimmingCharacters(in: .whitespacesAndNewlines)
         do {
-            let outcome = try await service.sendReport(isManual: true)
-            bugReportMessage = outcome.message
+            let outcome = try await service.sendReport(
+                isManual: true,
+                userDescription: description.isEmpty ? nil : description
+            )
+            if outcome.success {
+                bugReportPhase = .success(outcome.message)
+            } else {
+                bugReportPhase = .failure(outcome.message)
+            }
         } catch {
-            bugReportMessage = error.localizedDescription
+            bugReportPhase = .failure(error.localizedDescription)
+        }
+    }
+
+    private func openGitHubIssue() {
+        let description = bugReportUserDescription.trimmingCharacters(in: .whitespacesAndNewlines)
+
+        var bodyParts: [String] = []
+        bodyParts.append("## Describe the bug")
+        bodyParts.append("")
+        if !description.isEmpty {
+            bodyParts.append(description)
+        } else {
+            bodyParts.append("A clear and concise description of what the bug is.")
+        }
+        bodyParts.append("")
+        bodyParts.append("## Environment")
+        bodyParts.append("")
+        bodyParts.append("- macOS Version: \(ProcessInfo.processInfo.operatingSystemVersionString)")
+        bodyParts.append("- EXO Version: \(buildTag) (\(buildCommit))")
+        bodyParts.append("")
+        bodyParts.append("## Additional context")
+        bodyParts.append("")
+        bodyParts.append("A bug report with diagnostic logs was submitted via the app.")
+
+        let body = bodyParts.joined(separator: "\n")
+
+        var components = URLComponents(string: "https://github.com/exo-explore/exo/issues/new")!
+        components.queryItems = [
+            URLQueryItem(name: "template", value: "bug_report.md"),
+            URLQueryItem(name: "title", value: "[BUG] "),
+            URLQueryItem(name: "body", value: body),
+            URLQueryItem(name: "labels", value: "bug"),
+        ]
+
+        if let url = components.url {
+            NSWorkspace.shared.open(url)
         }
-        bugReportInFlight = false
     }
 
     private func showUninstallConfirmationAlert() {

app/EXO/EXO/EXOApp.swift

@@ -21,7 +21,9 @@ struct EXOApp: App {
     @StateObject private var localNetworkChecker: LocalNetworkChecker
     @StateObject private var updater: SparkleUpdater
     @StateObject private var thunderboltBridgeService: ThunderboltBridgeService
+    @StateObject private var settingsWindowController: SettingsWindowController
     private let terminationObserver: TerminationObserver
+    private let firstLaunchPopout = FirstLaunchPopout()
     private let ciContext = CIContext(options: nil)
 
     init() {
@@ -43,12 +45,13 @@ struct EXOApp: App {
         _updater = StateObject(wrappedValue: updater)
         let thunderboltBridge = ThunderboltBridgeService(clusterStateService: service)
         _thunderboltBridgeService = StateObject(wrappedValue: thunderboltBridge)
+        _settingsWindowController = StateObject(wrappedValue: SettingsWindowController())
         enableLaunchAtLoginIfNeeded()
         // Install LaunchDaemon to disable Thunderbolt Bridge on startup (prevents network loops)
         NetworkSetupHelper.promptAndInstallIfNeeded()
         // Check local network access periodically (warning disappears when user grants permission)
         localNetwork.startPeriodicChecking(interval: 10)
-        controller.scheduleLaunch(after: 15)
+        controller.scheduleLaunch(after: 5)
         service.startPolling()
         networkStatus.startPolling()
     }
@@ -62,8 +65,19 @@ struct EXOApp: App {
                 .environmentObject(localNetworkChecker)
                 .environmentObject(updater)
                 .environmentObject(thunderboltBridgeService)
+                .environmentObject(settingsWindowController)
         } label: {
             menuBarIcon
+                .onReceive(controller.$isFirstLaunchReady) { ready in
+                    if ready {
+                        DispatchQueue.main.asyncAfter(deadline: .now() + 3.0) {
+                            self.firstLaunchPopout.onComplete = { [weak controller] in
+                                controller?.markOnboardingCompleted()
+                            }
+                            self.firstLaunchPopout.show()
+                        }
+                    }
+                }
         }
         .menuBarExtraStyle(.window)
     }

app/EXO/EXO/ExoProcessController.swift

@@ -4,7 +4,10 @@ import Foundation
 
 private let customNamespaceKey = "EXOCustomNamespace"
 private let hfTokenKey = "EXOHFToken"
+private let hfEndpointKey = "EXOHFEndpoint"
 private let enableImageModelsKey = "EXOEnableImageModels"
+private let offlineModeKey = "EXOOfflineMode"
+private let onboardingCompletedKey = "EXOOnboardingCompleted"
 
 @MainActor
 final class ExoProcessController: ObservableObject {
@@ -51,6 +54,14 @@ final class ExoProcessController: ObservableObject {
             UserDefaults.standard.set(hfToken, forKey: hfTokenKey)
         }
     }
+    @Published var hfEndpoint: String = {
+        return UserDefaults.standard.string(forKey: hfEndpointKey) ?? ""
+    }()
+    {
+        didSet {
+            UserDefaults.standard.set(hfEndpoint, forKey: hfEndpointKey)
+        }
+    }
     @Published var enableImageModels: Bool = {
         return UserDefaults.standard.bool(forKey: enableImageModelsKey)
     }()
@@ -59,6 +70,17 @@ final class ExoProcessController: ObservableObject {
             UserDefaults.standard.set(enableImageModels, forKey: enableImageModelsKey)
         }
     }
+    @Published var offlineMode: Bool = {
+        return UserDefaults.standard.bool(forKey: offlineModeKey)
+    }()
+    {
+        didSet {
+            UserDefaults.standard.set(offlineMode, forKey: offlineModeKey)
+        }
+    }
+
+    /// Fires once when EXO transitions to `.running` for the very first time (fresh install).
+    @Published private(set) var isFirstLaunchReady = false
 
     private var process: Process?
     private var runtimeDirectoryURL: URL?
@@ -113,6 +135,9 @@ final class ExoProcessController: ObservableObject {
             try child.run()
             process = child
             status = .running
+
+            // Show welcome popout on every launch
+            isFirstLaunchReady = true
         } catch {
             process = nil
             status = .failed(message: "Launch error")
@@ -126,11 +151,37 @@ final class ExoProcessController: ObservableObject {
             return
         }
         process.terminationHandler = nil
-        if process.isRunning {
-            process.terminate()
-        }
-        self.process = nil
         status = .stopped
+
+        guard process.isRunning else {
+            self.process = nil
+            return
+        }
+
+        let proc = process
+        self.process = nil
+
+        Task.detached {
+            proc.interrupt()
+
+            for _ in 0..<50 {
+                if !proc.isRunning { return }
+                try? await Task.sleep(nanoseconds: 100_000_000)
+            }
+
+            if proc.isRunning {
+                proc.terminate()
+            }
+
+            for _ in 0..<30 {
+                if !proc.isRunning { return }
+                try? await Task.sleep(nanoseconds: 100_000_000)
+            }
+
+            if proc.isRunning {
+                kill(proc.processIdentifier, SIGKILL)
+            }
+        }
     }
 
     func restart() {
@@ -138,6 +189,17 @@ final class ExoProcessController: ObservableObject {
         launch()
     }
 
+    /// Mark onboarding as completed (user interacted with the welcome popout).
+    func markOnboardingCompleted() {
+        UserDefaults.standard.set(true, forKey: onboardingCompletedKey)
+    }
+
+    /// Reset onboarding so the welcome popout appears on next launch.
+    func resetOnboarding() {
+        UserDefaults.standard.removeObject(forKey: onboardingCompletedKey)
+        isFirstLaunchReady = false
+    }
+
     func scheduleLaunch(after seconds: TimeInterval) {
         cancelPendingLaunch()
         let start = max(1, Int(ceil(seconds)))
@@ -220,9 +282,15 @@ final class ExoProcessController: ObservableObject {
         if !hfToken.isEmpty {
             environment["HF_TOKEN"] = hfToken
         }
+        if !hfEndpoint.isEmpty {
+            environment["HF_ENDPOINT"] = hfEndpoint
+        }
         if enableImageModels {
             environment["EXO_ENABLE_IMAGE_MODELS"] = "true"
         }
+        if offlineMode {
+            environment["EXO_OFFLINE"] = "true"
+        }
 
         var paths: [String] = []
         if let existing = environment["PATH"], !existing.isEmpty {

app/EXO/EXO/Services/BugReportService.swift

@@ -38,7 +38,8 @@ struct BugReportService {
     func sendReport(
         baseURL: URL = URL(string: "http://127.0.0.1:52415")!,
         now: Date = Date(),
-        isManual: Bool = false
+        isManual: Bool = false,
+        userDescription: String? = nil
     ) async throws -> BugReportOutcome {
         let timestamp = Self.runTimestampString(now)
         let dayPrefix = Self.dayPrefixString(now)
@@ -60,7 +61,8 @@ struct BugReportService {
             ifconfig: ifconfigText,
             debugInfo: debugInfo,
             isManual: isManual,
-            clusterTbBridgeStatus: clusterTbBridgeStatus
+            clusterTbBridgeStatus: clusterTbBridgeStatus,
+            userDescription: userDescription
         )
 
         let eventLogFiles = readAllEventLogs()
@@ -306,7 +308,8 @@ struct BugReportService {
         ifconfig: String,
         debugInfo: DebugInfo,
         isManual: Bool,
-        clusterTbBridgeStatus: [[String: Any]]?
+        clusterTbBridgeStatus: [[String: Any]]?,
+        userDescription: String? = nil
     ) -> Data? {
         let system = readSystemMetadata()
         let exo = readExoMetadata()
@@ -323,6 +326,9 @@ struct BugReportService {
         if let tbStatus = clusterTbBridgeStatus {
             payload["cluster_thunderbolt_bridge"] = tbStatus
         }
+        if let desc = userDescription, !desc.isEmpty {
+            payload["user_description"] = desc
+        }
         return try? JSONSerialization.data(withJSONObject: payload, options: [.prettyPrinted])
     }
 

app/EXO/EXO/Views/FirstLaunchPopout.swift

@@ -0,0 +1,200 @@
+import AppKit
+import SwiftUI
+
+/// A popover callout anchored to the menu bar icon on every launch,
+/// pointing the user to the web dashboard with an arrow connecting to the icon.
+@MainActor
+final class FirstLaunchPopout {
+    private var popover: NSPopover?
+    private var countdownTask: Task<Void, Never>?
+    private static let dashboardURL = "http://localhost:52415/"
+
+    /// Called when the user completes onboarding (clicks Open Dashboard or dismisses).
+    var onComplete: (() -> Void)?
+
+    func show() {
+        guard popover == nil else { return }
+
+        // The status bar button may not exist yet on first launch; retry generously.
+        showWithRetry(attemptsRemaining: 15)
+    }
+
+    private func showWithRetry(attemptsRemaining: Int) {
+        guard attemptsRemaining > 0 else {
+            // Exhausted retries — fall back to just opening the dashboard directly.
+            openDashboard()
+            onComplete?()
+            return
+        }
+
+        guard let button = Self.findStatusItemButton() else {
+            DispatchQueue.main.asyncAfter(deadline: .now() + 1.0) { [weak self] in
+                self?.showWithRetry(attemptsRemaining: attemptsRemaining - 1)
+            }
+            return
+        }
+
+        let pop = NSPopover()
+        pop.behavior = .applicationDefined
+        pop.animates = true
+        pop.contentSize = NSSize(width: 280, height: 120)
+        pop.contentViewController = NSHostingController(
+            rootView: WelcomeCalloutView(
+                countdownDuration: 10,
+                onDismiss: { [weak self] in
+                    self?.onComplete?()
+                    self?.dismiss()
+                },
+                onOpen: { [weak self] in
+                    self?.openDashboard()
+                    self?.onComplete?()
+                    self?.dismiss()
+                }
+            )
+        )
+
+        self.popover = pop
+        pop.show(relativeTo: button.bounds, of: button, preferredEdge: .minY)
+
+        // Auto-open dashboard after 10s then dismiss
+        countdownTask = Task {
+            try? await Task.sleep(nanoseconds: 10_000_000_000)
+            if !Task.isCancelled {
+                openDashboard()
+                onComplete?()
+                dismiss()
+            }
+        }
+    }
+
+    func dismiss() {
+        countdownTask?.cancel()
+        countdownTask = nil
+        guard let pop = popover else { return }
+        popover = nil
+        pop.performClose(nil)
+    }
+
+    private func openDashboard() {
+        guard let url = URL(string: Self.dashboardURL) else { return }
+        NSWorkspace.shared.open(url)
+    }
+
+    /// Finds the NSStatusBarButton created by SwiftUI's MenuBarExtra.
+    /// Walks the view hierarchy to find the actual button rather than the content view.
+    private static func findStatusItemButton() -> NSView? {
+        for window in NSApp.windows {
+            let className = NSStringFromClass(type(of: window))
+            // Match NSStatusBarWindow or any internal SwiftUI status bar window
+            guard className.contains("StatusBar") || className.contains("MenuBarExtra") else {
+                continue
+            }
+            if let content = window.contentView {
+                if let button = findButton(in: content) {
+                    return button
+                }
+                // Fall back to the content view itself if it has a non-zero frame
+                if content.frame.width > 0 {
+                    return content
+                }
+            }
+        }
+        return nil
+    }
+
+    /// Recursively searches the view hierarchy for an NSStatusBarButton.
+    private static func findButton(in view: NSView) -> NSView? {
+        let className = NSStringFromClass(type(of: view))
+        if className.contains("StatusBarButton") || className.contains("StatusItem") {
+            return view
+        }
+        for subview in view.subviews {
+            if let found = findButton(in: subview) {
+                return found
+            }
+        }
+        return nil
+    }
+}
+
+/// Minimal welcome callout — friendly pointer, not a wall of text.
+/// Rendered inside the NSPopover which provides its own chrome and arrow.
+private struct WelcomeCalloutView: View {
+    let countdownDuration: Int
+    let onDismiss: () -> Void
+    let onOpen: () -> Void
+    @State private var countdown: Int
+    @State private var timerTask: Task<Void, Never>?
+
+    init(countdownDuration: Int, onDismiss: @escaping () -> Void, onOpen: @escaping () -> Void) {
+        self.countdownDuration = countdownDuration
+        self.onDismiss = onDismiss
+        self.onOpen = onOpen
+        self._countdown = State(initialValue: countdownDuration)
+    }
+
+    var body: some View {
+        VStack(alignment: .leading, spacing: 10) {
+            HStack(alignment: .top) {
+                Text("EXO is running")
+                    .font(.system(.headline, design: .rounded))
+                    .fontWeight(.semibold)
+                    .foregroundColor(.primary)
+                Spacer()
+                Button {
+                    onDismiss()
+                } label: {
+                    Image(systemName: "xmark.circle.fill")
+                        .font(.system(size: 14))
+                        .foregroundStyle(.tertiary)
+                }
+                .buttonStyle(.plain)
+            }
+
+            Text("Run your first model here:")
+                .font(.system(.subheadline, design: .default))
+                .foregroundColor(.secondary)
+
+            HStack {
+                Button {
+                    onOpen()
+                } label: {
+                    Label("Open Dashboard", systemImage: "arrow.up.right.square")
+                        .font(.system(.caption, design: .default))
+                        .fontWeight(.medium)
+                }
+                .buttonStyle(.borderedProminent)
+                .tint(.accentColor)
+                .controlSize(.small)
+
+                Spacer()
+
+                if countdown > 0 {
+                    Text("Launching in \(countdown) secs...")
+                        .font(.system(.caption2, design: .default))
+                        .foregroundColor(.secondary.opacity(0.6))
+                        .monospacedDigit()
+                }
+            }
+        }
+        .padding(14)
+        .onAppear {
+            startCountdown()
+        }
+        .onDisappear {
+            timerTask?.cancel()
+            timerTask = nil
+        }
+    }
+
+    private func startCountdown() {
+        timerTask = Task {
+            while countdown > 0 {
+                try? await Task.sleep(nanoseconds: 1_000_000_000)
+                if !Task.isCancelled {
+                    countdown -= 1
+                }
+            }
+        }
+    }
+}

app/EXO/EXO/Views/SettingsView.swift

@@ -0,0 +1,504 @@
+import AppKit
+import SwiftUI
+
+/// Native macOS Settings window following Apple HIG.
+/// Organized into General, Model, Advanced, and About sections.
+struct SettingsView: View {
+    @EnvironmentObject private var controller: ExoProcessController
+    @EnvironmentObject private var updater: SparkleUpdater
+    @EnvironmentObject private var networkStatusService: NetworkStatusService
+    @EnvironmentObject private var thunderboltBridgeService: ThunderboltBridgeService
+    @EnvironmentObject private var stateService: ClusterStateService
+
+    @State private var pendingNamespace: String = ""
+    @State private var pendingHFToken: String = ""
+    @State private var pendingHFEndpoint: String = ""
+    @State private var pendingEnableImageModels = false
+    @State private var pendingOfflineMode = false
+    @State private var needsRestart = false
+    @State private var bugReportInFlight = false
+    @State private var bugReportMessage: String?
+    @State private var uninstallInProgress = false
+
+    var body: some View {
+        TabView {
+            generalTab
+                .tabItem {
+                    Label("General", systemImage: "gear")
+                }
+            modelTab
+                .tabItem {
+                    Label("Model", systemImage: "cube")
+                }
+            advancedTab
+                .tabItem {
+                    Label("Advanced", systemImage: "wrench.and.screwdriver")
+                }
+            aboutTab
+                .tabItem {
+                    Label("About", systemImage: "info.circle")
+                }
+        }
+        .frame(width: 450, height: 400)
+        .onAppear {
+            pendingNamespace = controller.customNamespace
+            pendingHFToken = controller.hfToken
+            pendingHFEndpoint = controller.hfEndpoint
+            pendingEnableImageModels = controller.enableImageModels
+            pendingOfflineMode = controller.offlineMode
+            needsRestart = false
+        }
+    }
+
+    // MARK: - General Tab
+
+    private var generalTab: some View {
+        Form {
+            Section {
+                LabeledContent("Cluster Namespace") {
+                    TextField("default", text: $pendingNamespace)
+                        .textFieldStyle(.roundedBorder)
+                        .frame(width: 200)
+                }
+                Text("Nodes with the same namespace form a cluster. Leave empty for default.")
+                    .font(.caption)
+                    .foregroundColor(.secondary)
+            }
+
+            Section {
+                LabeledContent("HuggingFace Token") {
+                    SecureField("optional", text: $pendingHFToken)
+                        .textFieldStyle(.roundedBorder)
+                        .frame(width: 200)
+                }
+                Text("Required for gated models. Get yours at huggingface.co/settings/tokens")
+                    .font(.caption)
+                    .foregroundColor(.secondary)
+            }
+
+            Section {
+                LabeledContent("HuggingFace Endpoint") {
+                    TextField("default", text: $pendingHFEndpoint)
+                        .textFieldStyle(.roundedBorder)
+                        .frame(width: 200)
+                }
+                Text("Defaults to huggingface.co. Use a mirror (e.g. hf-mirror.com) for China.")
+                    .font(.caption)
+                    .foregroundColor(.secondary)
+            }
+
+            Section {
+                Toggle("Offline Mode", isOn: $pendingOfflineMode)
+                Text("Skip internet checks and use only locally available models.")
+                    .font(.caption)
+                    .foregroundColor(.secondary)
+            }
+
+            Section {
+                HStack {
+                    Spacer()
+                    Button("Save & Restart") {
+                        applyGeneralSettings()
+                    }
+                    .disabled(!hasGeneralChanges)
+                }
+            }
+        }
+        .formStyle(.grouped)
+        .padding()
+    }
+
+    // MARK: - Model Tab
+
+    private var modelTab: some View {
+        Form {
+            Section {
+                Toggle("Enable Image Models (experimental)", isOn: $pendingEnableImageModels)
+                Text("Allow text-to-image and image-to-image models in the model picker.")
+                    .font(.caption)
+                    .foregroundColor(.secondary)
+            }
+
+            Section {
+                HStack {
+                    Spacer()
+                    Button("Save & Restart") {
+                        applyModelSettings()
+                    }
+                    .disabled(!hasModelChanges)
+                }
+            }
+        }
+        .formStyle(.grouped)
+        .padding()
+    }
+
+    // MARK: - Advanced Tab
+
+    private var advancedTab: some View {
+        Form {
+            Section("Onboarding") {
+                HStack {
+                    VStack(alignment: .leading) {
+                        Text("Reset Onboarding")
+                        Text("Opens the dashboard and resets the onboarding wizard.")
+                            .font(.caption)
+                            .foregroundColor(.secondary)
+                    }
+                    Spacer()
+                    Button("Reset") {
+                        guard let url = URL(string: "http://localhost:52415/?reset-onboarding")
+                        else { return }
+                        NSWorkspace.shared.open(url)
+                    }
+                }
+            }
+
+            Section("Debug Info") {
+                LabeledContent("Thunderbolt Bridge") {
+                    Text(thunderboltStatusText)
+                        .foregroundColor(thunderboltStatusColor)
+                }
+
+                VStack(alignment: .leading, spacing: 2) {
+                    clusterThunderboltBridgeView
+                }
+
+                VStack(alignment: .leading, spacing: 2) {
+                    interfaceIpList
+                }
+
+                VStack(alignment: .leading, spacing: 2) {
+                    rdmaStatusView
+                }
+
+                sendBugReportButton
+            }
+
+            Section("Danger Zone") {
+                Button(role: .destructive) {
+                    showUninstallConfirmationAlert()
+                } label: {
+                    HStack {
+                        Text("Uninstall EXO")
+                        Spacer()
+                        Image(systemName: "trash")
+                            .imageScale(.small)
+                    }
+                }
+                .disabled(uninstallInProgress)
+            }
+        }
+        .formStyle(.grouped)
+        .padding()
+    }
+
+    // MARK: - About Tab
+
+    private var aboutTab: some View {
+        Form {
+            Section {
+                LabeledContent("Version") {
+                    Text(buildTag)
+                        .textSelection(.enabled)
+                }
+                LabeledContent("Commit") {
+                    Text(buildCommit)
+                        .font(.system(.body, design: .monospaced))
+                        .textSelection(.enabled)
+                }
+            }
+
+            Section {
+                Button("Check for Updates") {
+                    updater.checkForUpdates()
+                }
+            }
+        }
+        .formStyle(.grouped)
+        .padding()
+    }
+
+    // MARK: - Debug Info Views (moved from ContentView)
+
+    private var thunderboltStatusText: String {
+        switch networkStatusService.status.thunderboltBridgeState {
+        case .some(.disabled):
+            return "Disabled"
+        case .some(.deleted):
+            return "Deleted"
+        case .some(.enabled):
+            return "Enabled"
+        case nil:
+            return "Unknown"
+        }
+    }
+
+    private var thunderboltStatusColor: Color {
+        switch networkStatusService.status.thunderboltBridgeState {
+        case .some(.disabled), .some(.deleted):
+            return .green
+        case .some(.enabled):
+            return .red
+        case nil:
+            return .secondary
+        }
+    }
+
+    private var clusterThunderboltBridgeView: some View {
+        let bridgeStatuses = stateService.latestSnapshot?.nodeThunderboltBridge ?? [:]
+        let localNodeId = stateService.localNodeId
+        let nodeProfiles = stateService.latestSnapshot?.nodeProfiles ?? [:]
+
+        return VStack(alignment: .leading, spacing: 1) {
+            if bridgeStatuses.isEmpty {
+                Text("Cluster TB Bridge: No data")
+                    .font(.caption2)
+                    .foregroundColor(.secondary)
+            } else {
+                Text("Cluster TB Bridge Status:")
+                    .font(.caption2)
+                    .foregroundColor(.secondary)
+                ForEach(Array(bridgeStatuses.keys.sorted()), id: \.self) { nodeId in
+                    if let status = bridgeStatuses[nodeId] {
+                        let nodeName =
+                            nodeProfiles[nodeId]?.friendlyName ?? String(nodeId.prefix(8))
+                        let isLocal = nodeId == localNodeId
+                        let prefix = isLocal ? "  \(nodeName) (local):" : "  \(nodeName):"
+                        let statusText =
+                            !status.exists
+                            ? "N/A"
+                            : (status.enabled ? "Enabled" : "Disabled")
+                        let color: Color =
+                            !status.exists
+                            ? .secondary
+                            : (status.enabled ? .red : .green)
+                        Text("\(prefix) \(statusText)")
+                            .font(.caption2)
+                            .foregroundColor(color)
+                    }
+                }
+            }
+        }
+    }
+
+    private var interfaceIpList: some View {
+        let statuses = networkStatusService.status.interfaceStatuses
+        return VStack(alignment: .leading, spacing: 1) {
+            Text("Interfaces (en0–en7):")
+                .font(.caption2)
+                .foregroundColor(.secondary)
+            if statuses.isEmpty {
+                Text("  Unknown")
+                    .font(.caption2)
+                    .foregroundColor(.secondary)
+            } else {
+                ForEach(statuses, id: \.interfaceName) { status in
+                    let ipText = status.ipAddress ?? "No IP"
+                    Text("  \(status.interfaceName): \(ipText)")
+                        .font(.caption2)
+                        .foregroundColor(status.ipAddress == nil ? .red : .green)
+                }
+            }
+        }
+    }
+
+    private var rdmaStatusView: some View {
+        let rdmaStatuses = stateService.latestSnapshot?.nodeRdmaCtl ?? [:]
+        let localNodeId = stateService.localNodeId
+        let nodeProfiles = stateService.latestSnapshot?.nodeProfiles ?? [:]
+        let localDevices = networkStatusService.status.localRdmaDevices
+        let localPorts = networkStatusService.status.localRdmaActivePorts
+
+        return VStack(alignment: .leading, spacing: 1) {
+            if rdmaStatuses.isEmpty {
+                Text("Cluster RDMA: No data")
+                    .font(.caption2)
+                    .foregroundColor(.secondary)
+            } else {
+                Text("Cluster RDMA Status:")
+                    .font(.caption2)
+                    .foregroundColor(.secondary)
+                ForEach(Array(rdmaStatuses.keys.sorted()), id: \.self) { nodeId in
+                    if let status = rdmaStatuses[nodeId] {
+                        let nodeName =
+                            nodeProfiles[nodeId]?.friendlyName ?? String(nodeId.prefix(8))
+                        let isLocal = nodeId == localNodeId
+                        let prefix = isLocal ? "  \(nodeName) (local):" : "  \(nodeName):"
+                        let statusText = status.enabled ? "Enabled" : "Disabled"
+                        let color: Color = status.enabled ? .green : .orange
+                        Text("\(prefix) \(statusText)")
+                            .font(.caption2)
+                            .foregroundColor(color)
+                    }
+                }
+            }
+            if !localDevices.isEmpty {
+                Text("  Local Devices: \(localDevices.joined(separator: ", "))")
+                    .font(.caption2)
+                    .foregroundColor(.secondary)
+            }
+            if !localPorts.isEmpty {
+                Text("  Local Active Ports:")
+                    .font(.caption2)
+                    .foregroundColor(.secondary)
+                ForEach(localPorts, id: \.device) { port in
+                    Text("    \(port.device) port \(port.port): \(port.state)")
+                        .font(.caption2)
+                        .foregroundColor(.green)
+                }
+            }
+        }
+    }
+
+    private var sendBugReportButton: some View {
+        VStack(alignment: .leading, spacing: 4) {
+            Button {
+                Task {
+                    await sendBugReport()
+                }
+            } label: {
+                HStack {
+                    if bugReportInFlight {
+                        ProgressView()
+                            .scaleEffect(0.6)
+                    }
+                    Text("Send Bug Report")
+                        .font(.caption)
+                        .fontWeight(.semibold)
+                    Spacer()
+                }
+            }
+            .disabled(bugReportInFlight)
+
+            if let message = bugReportMessage {
+                Text(message)
+                    .font(.caption2)
+                    .foregroundColor(.secondary)
+                    .fixedSize(horizontal: false, vertical: true)
+            }
+        }
+    }
+
+    // MARK: - Actions
+
+    private func sendBugReport() async {
+        bugReportInFlight = true
+        bugReportMessage = "Collecting logs..."
+        let service = BugReportService()
+        do {
+            let outcome = try await service.sendReport(isManual: true)
+            bugReportMessage = outcome.message
+        } catch {
+            bugReportMessage = error.localizedDescription
+        }
+        bugReportInFlight = false
+    }
+
+    private func showUninstallConfirmationAlert() {
+        let alert = NSAlert()
+        alert.messageText = "Uninstall EXO"
+        alert.informativeText = """
+            This will remove EXO and all its system components:
+
+            • Network configuration daemon
+            • Launch at login registration
+            • EXO network location
+
+            The app will be moved to Trash.
+            """
+        alert.alertStyle = .warning
+        alert.addButton(withTitle: "Uninstall")
+        alert.addButton(withTitle: "Cancel")
+
+        if let uninstallButton = alert.buttons.first {
+            uninstallButton.hasDestructiveAction = true
+        }
+
+        let response = alert.runModal()
+        if response == .alertFirstButtonReturn {
+            performUninstall()
+        }
+    }
+
+    private func performUninstall() {
+        uninstallInProgress = true
+
+        controller.cancelPendingLaunch()
+        controller.stop()
+        stateService.stopPolling()
+
+        DispatchQueue.global(qos: .utility).async {
+            do {
+                try NetworkSetupHelper.uninstall()
+
+                DispatchQueue.main.async {
+                    LaunchAtLoginHelper.disable()
+                    self.moveAppToTrash()
+
+                    DispatchQueue.main.asyncAfter(deadline: .now() + 0.5) {
+                        NSApplication.shared.terminate(nil)
+                    }
+                }
+            } catch {
+                DispatchQueue.main.async {
+                    let errorAlert = NSAlert()
+                    errorAlert.messageText = "Uninstall Failed"
+                    errorAlert.informativeText = error.localizedDescription
+                    errorAlert.alertStyle = .critical
+                    errorAlert.addButton(withTitle: "OK")
+                    errorAlert.runModal()
+                    self.uninstallInProgress = false
+                }
+            }
+        }
+    }
+
+    private func moveAppToTrash() {
+        guard let appURL = Bundle.main.bundleURL as URL? else { return }
+        do {
+            try FileManager.default.trashItem(at: appURL, resultingItemURL: nil)
+        } catch {
+            // If we can't trash the app, that's OK - user can do it manually
+        }
+    }
+
+    // MARK: - Helpers
+
+    private var hasGeneralChanges: Bool {
+        pendingNamespace != controller.customNamespace || pendingHFToken != controller.hfToken
+            || pendingHFEndpoint != controller.hfEndpoint
+            || pendingOfflineMode != controller.offlineMode
+    }
+
+    private var hasModelChanges: Bool {
+        pendingEnableImageModels != controller.enableImageModels
+    }
+
+    private func applyGeneralSettings() {
+        controller.customNamespace = pendingNamespace
+        controller.hfToken = pendingHFToken
+        controller.hfEndpoint = pendingHFEndpoint
+        controller.offlineMode = pendingOfflineMode
+        restartIfRunning()
+    }
+
+    private func applyModelSettings() {
+        controller.enableImageModels = pendingEnableImageModels
+        restartIfRunning()
+    }
+
+    private func restartIfRunning() {
+        if controller.status == .running || controller.status == .starting {
+            controller.restart()
+        }
+    }
+
+    private var buildTag: String {
+        Bundle.main.infoDictionary?["EXOBuildTag"] as? String ?? "unknown"
+    }
+
+    private var buildCommit: String {
+        Bundle.main.infoDictionary?["EXOBuildCommit"] as? String ?? "unknown"
+    }
+}

app/EXO/EXO/Views/SettingsWindowController.swift

@@ -0,0 +1,47 @@
+import AppKit
+import SwiftUI
+
+/// Manages a standalone native macOS Settings window.
+/// Ensures only one instance exists and brings it to front on repeated opens.
+@MainActor
+final class SettingsWindowController: ObservableObject {
+    private var window: NSWindow?
+
+    func open(
+        controller: ExoProcessController,
+        updater: SparkleUpdater,
+        networkStatusService: NetworkStatusService,
+        thunderboltBridgeService: ThunderboltBridgeService,
+        stateService: ClusterStateService
+    ) {
+        if let existing = window, existing.isVisible {
+            existing.makeKeyAndOrderFront(nil)
+            NSApp.activate(ignoringOtherApps: true)
+            return
+        }
+
+        let settingsView = SettingsView()
+            .environmentObject(controller)
+            .environmentObject(updater)
+            .environmentObject(networkStatusService)
+            .environmentObject(thunderboltBridgeService)
+            .environmentObject(stateService)
+
+        let hostingView = NSHostingView(rootView: settingsView)
+
+        let newWindow = NSWindow(
+            contentRect: NSRect(x: 0, y: 0, width: 450, height: 400),
+            styleMask: [.titled, .closable],
+            backing: .buffered,
+            defer: false
+        )
+        newWindow.title = "EXO Settings"
+        newWindow.contentView = hostingView
+        newWindow.center()
+        newWindow.isReleasedWhenClosed = false
+        newWindow.makeKeyAndOrderFront(nil)
+        NSApp.activate(ignoringOtherApps: true)
+
+        window = newWindow
+    }
+}

bench/bench.toml

@@ -2,6 +2,4 @@
 #
 # Lists the suite files to include. Each file defines benchmarks
 # with shared constraints, topology, and default args.
-include = [
-    "single-m3-ultra.toml",
-]
+include = ["single-m3-ultra.toml"]

bench/eval_configs/models.toml

@@ -0,0 +1,292 @@
+# Model evaluation configurations for exo_eval.
+#
+# Each [[model]] entry uses `patterns` — a list of substrings matched
+# against the model_id. First matching entry wins.
+#
+# Required fields:
+#   name, patterns, reasoning
+#
+# Optional per-model overrides (CLI flags take priority over these):
+#   temperature, top_p, max_tokens, reasoning_effort
+#
+# Fallback defaults (when no per-model config):
+#   reasoning:     temperature=1.0, max_tokens=131072, reasoning_effort="high"
+#   non-reasoning: temperature=0.0, max_tokens=16384
+#
+# All per-model values below are sourced from official model cards,
+# generation_config.json files, and vendor documentation.
+
+# ─── Qwen3.5 (Feb 2026) ─────────────────────────────────────────────
+# Source: HuggingFace model cards (Qwen/Qwen3.5-*)
+# 35B-A3B thinking general: temp=1.0, top_p=0.95, top_k=20
+# 397B thinking: temp=0.6, top_p=0.95, top_k=20
+# Non-thinking: temp=0.7, top_p=0.8, top_k=20
+# max_tokens: 32768 general, 81920 for complex math/code
+
+[[model]]
+name = "Qwen3.5 2B"
+patterns = ["Qwen3.5-2B"]
+reasoning = true
+temperature = 0.6
+top_p = 0.95
+max_tokens = 81920
+
+[[model]]
+name = "Qwen3.5 9B"
+patterns = ["Qwen3.5-9B"]
+reasoning = true
+temperature = 0.6
+top_p = 0.95
+max_tokens = 81920
+
+[[model]]
+name = "Qwen3.5 27B"
+patterns = ["Qwen3.5-27B"]
+reasoning = true
+temperature = 0.6
+top_p = 0.95
+max_tokens = 81920
+
+[[model]]
+name = "Qwen3.5 35B A3B"
+patterns = ["Qwen3.5-35B-A3B"]
+reasoning = true
+temperature = 1.0
+top_p = 0.95
+max_tokens = 81920
+
+[[model]]
+name = "Qwen3.5 122B A10B"
+patterns = ["Qwen3.5-122B-A10B"]
+reasoning = true
+temperature = 0.6
+top_p = 0.95
+max_tokens = 81920
+
+[[model]]
+name = "Qwen3.5 397B A17B"
+patterns = ["Qwen3.5-397B-A17B"]
+reasoning = true
+temperature = 0.6
+top_p = 0.95
+max_tokens = 81920
+
+# ─── Qwen3 (Apr 2025) ───────────────────────────────────────────────
+# Source: HuggingFace model cards (Qwen/Qwen3-*)
+# Thinking: temp=0.6, top_p=0.95, top_k=20
+# Non-thinking: temp=0.7, top_p=0.8, top_k=20
+# max_tokens: 32768 general, 38912 for complex math/code
+
+[[model]]
+name = "Qwen3 0.6B"
+patterns = ["Qwen3-0.6B"]
+reasoning = true
+temperature = 0.6
+top_p = 0.95
+max_tokens = 38912
+
+[[model]]
+name = "Qwen3 30B A3B"
+patterns = ["Qwen3-30B-A3B"]
+reasoning = true
+temperature = 0.6
+top_p = 0.95
+max_tokens = 38912
+
+[[model]]
+name = "Qwen3 235B A22B"
+patterns = ["Qwen3-235B-A22B"]
+reasoning = true
+temperature = 0.6
+top_p = 0.95
+max_tokens = 38912
+
+[[model]]
+name = "Qwen3 Next 80B Thinking"
+patterns = ["Qwen3-Next-80B-A3B-Thinking"]
+reasoning = true
+temperature = 0.6
+top_p = 0.95
+max_tokens = 38912
+
+[[model]]
+name = "Qwen3 Next 80B Instruct"
+patterns = ["Qwen3-Next-80B-A3B-Instruct"]
+reasoning = false
+temperature = 0.7
+top_p = 0.8
+max_tokens = 16384
+
+[[model]]
+name = "Qwen3 Coder 480B"
+patterns = ["Qwen3-Coder-480B"]
+reasoning = false
+temperature = 0.7
+top_p = 0.8
+max_tokens = 16384
+
+[[model]]
+name = "Qwen3 Coder Next"
+patterns = ["Qwen3-Coder-Next"]
+reasoning = false
+temperature = 0.7
+top_p = 0.8
+max_tokens = 16384
+
+# ─── GPT-OSS (OpenAI) ───────────────────────────────────────────────
+# Source: OpenAI GitHub README + HuggingFace discussion #21
+# temp=1.0, top_p=1.0, NO top_k, NO repetition_penalty
+# reasoning_effort supported: low/medium/high
+
+[[model]]
+name = "GPT-OSS 20B"
+patterns = ["gpt-oss-20b"]
+reasoning = true
+temperature = 1.0
+top_p = 1.0
+
+[[model]]
+name = "GPT-OSS 120B"
+patterns = ["gpt-oss-120b"]
+reasoning = true
+temperature = 1.0
+top_p = 1.0
+
+# ─── DeepSeek ────────────────────────────────────────────────────────
+# Source: https://api-docs.deepseek.com/quick_start/parameter_settings
+# Coding/Math: temp=0.0, General: temp=1.3, Creative: temp=1.5
+# NOTE: DeepSeek API applies nonlinear temp mapping. These are API values.
+# When running model directly: API temp 1.0 = model temp 0.3
+# We use temp=0.0 for eval (coding/math focus).
+
+[[model]]
+name = "DeepSeek V3.1"
+patterns = ["DeepSeek-V3.1"]
+reasoning = true
+temperature = 0.0
+
+# ─── GLM (ZhipuAI / THUDM) ──────────────────────────────────────────
+# Source: HuggingFace model cards + generation_config.json + docs.z.ai
+# GLM 4.5+: temp=1.0, top_p=0.95
+# Reasoning tasks: 131072 max_tokens; coding/SWE tasks: temp=0.7
+
+[[model]]
+name = "GLM-5"
+patterns = ["GLM-5"]
+reasoning = true
+temperature = 1.0
+top_p = 0.95
+max_tokens = 131072
+
+[[model]]
+name = "GLM 4.5 Air"
+patterns = ["GLM-4.5-Air"]
+reasoning = true
+temperature = 1.0
+top_p = 0.95
+
+[[model]]
+name = "GLM 4.7"
+patterns = ["GLM-4.7-"]
+reasoning = true
+temperature = 1.0
+top_p = 0.95
+max_tokens = 131072
+# Note: matches both GLM-4.7 and GLM-4.7-Flash
+
+# ─── Kimi (Moonshot AI) ─────────────────────────────────────────────
+# Source: HuggingFace model cards (moonshotai/Kimi-K2-*)
+# K2-Instruct: temp=0.6
+# K2-Thinking: temp=1.0, max_length=262144
+# K2.5: thinking temp=1.0, top_p=0.95; instant temp=0.6, top_p=0.95
+
+[[model]]
+name = "Kimi K2 Thinking"
+patterns = ["Kimi-K2-Thinking"]
+reasoning = true
+temperature = 1.0
+max_tokens = 131072
+
+[[model]]
+name = "Kimi K2.5"
+patterns = ["Kimi-K2.5"]
+reasoning = true
+temperature = 1.0
+top_p = 0.95
+max_tokens = 131072
+
+[[model]]
+name = "Kimi K2 Instruct"
+patterns = ["Kimi-K2-Instruct"]
+reasoning = false
+temperature = 0.6
+
+# ─── MiniMax ─────────────────────────────────────────────────────────
+# Source: HuggingFace model cards + generation_config.json
+# All models: temp=1.0, top_p=0.95, top_k=40
+
+[[model]]
+name = "MiniMax M2.5"
+patterns = ["MiniMax-M2.5"]
+reasoning = true
+temperature = 1.0
+top_p = 0.95
+
+[[model]]
+name = "MiniMax M2.1"
+patterns = ["MiniMax-M2.1"]
+reasoning = true
+temperature = 1.0
+top_p = 0.95
+
+# ─── Step (StepFun) ─────────────────────────────────────────────────
+# Source: HuggingFace model card (stepfun-ai/Step-3.5-Flash)
+# Reasoning: temp=1.0, top_p=0.95
+# General chat: temp=0.6, top_p=0.95
+# We use reasoning settings for eval.
+
+[[model]]
+name = "Step 3.5 Flash"
+patterns = ["Step-3.5-Flash"]
+reasoning = true
+temperature = 1.0
+top_p = 0.95
+
+# ─── Llama (Meta) ───────────────────────────────────────────────────
+# Source: generation_config.json + meta-llama/llama-models generation.py
+# All variants: temp=0.6, top_p=0.9
+
+[[model]]
+name = "Llama 3.2 1B"
+patterns = ["Llama-3.2-1B"]
+reasoning = false
+temperature = 0.6
+top_p = 0.9
+
+[[model]]
+name = "Llama 3.2 3B"
+patterns = ["Llama-3.2-3B"]
+reasoning = false
+temperature = 0.6
+top_p = 0.9
+
+[[model]]
+name = "Llama 3.1 8B"
+patterns = ["Llama-3.1-8B", "Meta-Llama-3.1-8B"]
+reasoning = false
+temperature = 0.6
+top_p = 0.9
+
+[[model]]
+name = "Llama 3.1 70B"
+patterns = ["Llama-3.1-70B", "Meta-Llama-3.1-70B"]
+reasoning = false
+temperature = 0.6
+top_p = 0.9
+
+[[model]]
+name = "Llama 3.3 70B"
+patterns = ["Llama-3.3-70B", "llama-3.3-70b"]
+reasoning = false
+temperature = 0.6
+top_p = 0.9

bench/exo_bench.py

@@ -1,29 +1,50 @@
+# type: ignore
 #!/usr/bin/env python3
-# pyright: reportAny=false, reportUnknownMemberType=false, reportUnknownVariableType=false, reportUnknownArgumentType=false
+"""Tool-calling eval for exo's OpenAI-compatible API.
+
+Tests whether models correctly:
+- Trigger tool calls when appropriate
+- Return valid JSON arguments matching function schemas
+- Handle multi-turn tool use (call -> result -> final answer)
+- Avoid calling tools when unnecessary
+
+Start exo with a model first, then run:
+    uv run python tool_call_eval.py --model <model-id>
+    uv run python tool_call_eval.py --model <model-id> --host 10.0.0.5 --port 52415
+    uv run python tool_call_eval.py --model <model-id> --repeat 3
+    uv run python tool_call_eval.py --model <model-id> --scenarios weather_simple calculator_multi_turn
+"""
+
 from __future__ import annotations
 
 import argparse
 import contextlib
-import http.client
 import itertools
 import json
-import os
 import sys
+import threading
 import time
 from collections.abc import Callable
+from concurrent.futures import ThreadPoolExecutor, as_completed
 from pathlib import Path
 from statistics import mean
 from typing import Any
-from urllib.parse import urlencode
 
+from harness import (
+    ExoClient,
+    ExoHttpError,
+    add_common_instance_args,
+    instance_id_from_instance,
+    nodes_used_in_instance,
+    resolve_model_short_id,
+    run_planning_phase,
+    settle_and_fetch_placements,
+    wait_for_instance_gone,
+    wait_for_instance_ready,
+)
 from loguru import logger
 from transformers import AutoTokenizer
 
-# Backoff constants for cluster settling retry
-_SETTLE_INITIAL_BACKOFF_S = 1.0
-_SETTLE_MAX_BACKOFF_S = 60.0
-_SETTLE_BACKOFF_MULTIPLIER = 2.0
-
 # Monkey-patch for transformers 5.x compatibility
 # Kimi's tokenization_kimi.py imports bytes_to_unicode from the old location
 # which was moved in transformers 5.0.0rc2
@@ -56,7 +77,7 @@ def load_tokenizer_for_bench(model_id: str) -> Any:
         model_path = Path(
             snapshot_download(
                 model_id,
-                allow_patterns=["*.json", "*.py", "*.tiktoken"],
+                allow_patterns=["*.json", "*.py", "*.tiktoken", "*.model"],
             )
         )
 
@@ -103,154 +124,6 @@ def load_tokenizer_for_bench(model_id: str) -> Any:
     return AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
 
 
-class ExoHttpError(RuntimeError):
-    def __init__(self, status: int, reason: str, body_preview: str):
-        super().__init__(f"HTTP {status} {reason}: {body_preview}")
-        self.status = status
-
-
-class ExoClient:
-    def __init__(self, host: str, port: int, timeout_s: float = 7200.0):
-        self.host = host
-        self.port = port
-        self.timeout_s = timeout_s
-
-    def request_json(
-        self,
-        method: str,
-        path: str,
-        params: dict[str, Any] | None = None,
-        body: dict[str, Any] | None = None,
-        headers: dict[str, str] | None = None,
-    ) -> Any:
-        if not path.startswith("/"):
-            path = "/" + path
-        if params:
-            path = path + "?" + urlencode(params)
-
-        conn = http.client.HTTPConnection(self.host, self.port, timeout=self.timeout_s)
-        try:
-            payload: bytes | None = None
-            hdrs: dict[str, str] = {"Accept": "application/json"}
-
-            if body is not None:
-                payload = json.dumps(body).encode("utf-8")
-                hdrs["Content-Type"] = "application/json"
-            if headers:
-                hdrs.update(headers)
-
-            conn.request(method.upper(), path, body=payload, headers=hdrs)
-            resp = conn.getresponse()
-            raw = resp.read()
-            text = raw.decode("utf-8", errors="replace") if raw else ""
-
-            if resp.status >= 400:
-                raise ExoHttpError(resp.status, resp.reason, text[:300])
-
-            if not text:
-                return None
-            return json.loads(text)
-        finally:
-            conn.close()
-
-    def post_bench_chat_completions(self, payload: dict[str, Any]) -> dict[str, Any]:
-        return self.request_json("POST", "/bench/chat/completions", body=payload)
-
-
-def unwrap_instance(instance: dict[str, Any]) -> dict[str, Any]:
-    if len(instance) != 1:
-        raise KeyError(f"Expected 1 key, got keys={list(instance.keys())}")
-
-    tag = next(iter(instance))
-    inner = instance[tag]
-    if not isinstance(inner, dict):
-        raise TypeError(f"payload for {tag} must be dict, got {type(inner)}")
-    return inner
-
-
-def instance_id_from_instance(instance: dict[str, Any]) -> str:
-    inner = unwrap_instance(instance)
-    return str(inner["instanceId"])
-
-
-def nodes_used_in_instance(instance: dict[str, Any]) -> int:
-    inner = unwrap_instance(instance)
-    return len(inner["shardAssignments"]["nodeToRunner"])
-
-
-def runner_ids_from_instance(instance: dict[str, Any]) -> list[str]:
-    inner = unwrap_instance(instance)
-    runner_to_shard = inner["shardAssignments"]["runnerToShard"]
-    return list(runner_to_shard.keys())
-
-
-def runner_ready(runner: dict[str, Any]) -> bool:
-    return "RunnerReady" in runner
-
-
-def runner_failed(runner: dict[str, Any]) -> bool:
-    return "RunnerFailed" in runner
-
-
-def get_runner_failed_message(runner: dict[str, Any]) -> str | None:
-    if "RunnerFailed" in runner:
-        return runner["RunnerFailed"].get("errorMessage")
-    return None
-
-
-def wait_for_instance_ready(
-    client: ExoClient, instance_id: str, timeout: float = 24000.0
-) -> None:
-    start_time = time.time()
-    instance_existed = False
-    while time.time() - start_time < timeout:
-        state = client.request_json("GET", "/state")
-        instances = state.get("instances", {})
-
-        if instance_id not in instances:
-            if instance_existed:
-                # Instance was deleted after being created - likely due to runner failure
-                raise RuntimeError(
-                    f"Instance {instance_id} was deleted (runner may have failed)"
-                )
-            time.sleep(0.1)
-            continue
-
-        instance_existed = True
-        instance = instances[instance_id]
-        runner_ids = runner_ids_from_instance(instance)
-        runners = state.get("runners", {})
-
-        # Check for failed runners first
-        for rid in runner_ids:
-            runner = runners.get(rid, {})
-            if runner_failed(runner):
-                error_msg = get_runner_failed_message(runner) or "Unknown error"
-                raise RuntimeError(f"Runner {rid} failed: {error_msg}")
-
-        if all(runner_ready(runners.get(rid, {})) for rid in runner_ids):
-            return
-
-        time.sleep(0.1)
-
-    raise TimeoutError(f"Instance {instance_id} did not become ready within {timeout=}")
-
-
-def wait_for_instance_gone(
-    client: ExoClient, instance_id: str, timeout: float = 3.0
-) -> None:
-    start_time = time.time()
-    while time.time() - start_time < timeout:
-        try:
-            client.request_json("GET", f"/instance/{instance_id}")
-            time.sleep(0.4)
-        except ExoHttpError as e:
-            if e.status == 404:
-                return
-
-    raise TimeoutError(f"Instance {instance_id} did not get deleted within {timeout=}")
-
-
 def format_peak_memory(b: float) -> str:
     for unit in ["B", "KB", "MB", "GB", "TB"]:
         if b < 1024.0:
@@ -269,184 +142,6 @@ def parse_int_list(values: list[str]) -> list[int]:
     return items
 
 
-def resolve_model_short_id(client: ExoClient, model_arg: str) -> tuple[str, str]:
-    models = client.request_json("GET", "/models") or {}
-    data = models.get("data") or []
-
-    for m in data:
-        if m.get("name").lower() == model_arg.lower():
-            short_id = str(m["name"])
-            full_id = str(m.get("hugging_face_id") or m["name"])
-            return short_id, full_id
-
-    for m in data:
-        if m.get("hugging_face_id") == model_arg:
-            short_id = str(m["name"])
-            full_id = str(m["hugging_face_id"])
-            return short_id, full_id
-
-    raise ValueError(f"Model not found in /models: {model_arg}")
-
-
-def run_planning_phase(
-    client: ExoClient,
-    full_model_id: str,
-    preview: dict[str, Any],
-    danger_delete: bool,
-    timeout: float,
-    settle_deadline: float | None,
-) -> None:
-    """Check disk space and ensure model is downloaded before benchmarking."""
-    # Get model size from /models
-    models = client.request_json("GET", "/models") or {}
-    model_bytes = 0
-    for m in models.get("data", []):
-        if m.get("hugging_face_id") == full_model_id:
-            model_bytes = m.get("storage_size_megabytes", 0) * 1024 * 1024
-            break
-
-    if not model_bytes:
-        logger.warning(
-            f"Could not determine size for {full_model_id}, skipping disk check"
-        )
-        return
-
-    # Get nodes from preview
-    inner = unwrap_instance(preview["instance"])
-    node_ids = list(inner["shardAssignments"]["nodeToRunner"].keys())
-    runner_to_shard = inner["shardAssignments"]["runnerToShard"]
-
-    state = client.request_json("GET", "/state")
-    downloads = state.get("downloads", {})
-    node_disk = state.get("nodeDisk", {})
-
-    for node_id in node_ids:
-        node_downloads = downloads.get(node_id, [])
-
-        # Check if model already downloaded on this node
-        already_downloaded = any(
-            "DownloadCompleted" in p
-            and unwrap_instance(p["DownloadCompleted"]["shardMetadata"])["modelCard"][
-                "modelId"
-            ]
-            == full_model_id
-            for p in node_downloads
-        )
-        if already_downloaded:
-            continue
-
-        # Wait for disk info if settle_deadline is set
-        disk_info = node_disk.get(node_id, {})
-        backoff = _SETTLE_INITIAL_BACKOFF_S
-        while not disk_info and settle_deadline and time.monotonic() < settle_deadline:
-            remaining = settle_deadline - time.monotonic()
-            logger.info(
-                f"Waiting for disk info on {node_id} ({remaining:.0f}s remaining)..."
-            )
-            time.sleep(min(backoff, remaining))
-            backoff = min(backoff * _SETTLE_BACKOFF_MULTIPLIER, _SETTLE_MAX_BACKOFF_S)
-            state = client.request_json("GET", "/state")
-            node_disk = state.get("nodeDisk", {})
-            disk_info = node_disk.get(node_id, {})
-
-        if not disk_info:
-            logger.warning(f"No disk info for {node_id}, skipping space check")
-            continue
-
-        avail = disk_info.get("available", {}).get("inBytes", 0)
-        if avail >= model_bytes:
-            continue
-
-        if not danger_delete:
-            raise RuntimeError(
-                f"Insufficient disk on {node_id}: need {model_bytes // (1024**3)}GB, "
-                f"have {avail // (1024**3)}GB. Use --danger-delete-downloads to free space."
-            )
-
-        # Delete from smallest to largest
-        completed = [
-            (
-                unwrap_instance(p["DownloadCompleted"]["shardMetadata"])["modelCard"][
-                    "modelId"
-                ],
-                p["DownloadCompleted"]["totalBytes"]["inBytes"],
-            )
-            for p in node_downloads
-            if "DownloadCompleted" in p
-        ]
-        for del_model, size in sorted(completed, key=lambda x: x[1]):
-            logger.info(f"Deleting {del_model} from {node_id} ({size // (1024**2)}MB)")
-            client.request_json("DELETE", f"/download/{node_id}/{del_model}")
-            avail += size
-            if avail >= model_bytes:
-                break
-
-        if avail < model_bytes:
-            raise RuntimeError(f"Could not free enough space on {node_id}")
-
-    # Start downloads (idempotent)
-    for node_id in node_ids:
-        runner_id = inner["shardAssignments"]["nodeToRunner"][node_id]
-        shard = runner_to_shard[runner_id]
-        client.request_json(
-            "POST",
-            "/download/start",
-            body={
-                "targetNodeId": node_id,
-                "shardMetadata": shard,
-            },
-        )
-        logger.info(f"Started download on {node_id}")
-
-    # Wait for downloads
-    start = time.time()
-    while time.time() - start < timeout:
-        state = client.request_json("GET", "/state")
-        downloads = state.get("downloads", {})
-        all_done = True
-        for node_id in node_ids:
-            done = any(
-                "DownloadCompleted" in p
-                and unwrap_instance(p["DownloadCompleted"]["shardMetadata"])[
-                    "modelCard"
-                ]["modelId"]
-                == full_model_id
-                for p in downloads.get(node_id, [])
-            )
-            failed = [
-                p["DownloadFailed"]["errorMessage"]
-                for p in downloads.get(node_id, [])
-                if "DownloadFailed" in p
-                and unwrap_instance(p["DownloadFailed"]["shardMetadata"])["modelCard"][
-                    "modelId"
-                ]
-                == full_model_id
-            ]
-            if failed:
-                raise RuntimeError(f"Download failed on {node_id}: {failed[0]}")
-            if not done:
-                all_done = False
-        if all_done:
-            return
-        time.sleep(1)
-
-    raise TimeoutError("Downloads did not complete in time")
-
-
-def placement_filter(instance_meta: str, wanted: str) -> bool:
-    s = (instance_meta or "").lower()
-    if wanted == "both":
-        return ("ring" in s) or ("jaccl" in s)
-    return wanted in s
-
-
-def sharding_filter(sharding: str, wanted: str) -> bool:
-    s = (sharding or "").lower()
-    if wanted == "both":
-        return ("pipeline" in s) or ("tensor" in s)
-    return wanted in s
-
-
 def run_one_completion(
     client: ExoClient, model_id: str, pp_hint: int, tg: int, prompt_sizer: PromptSizer
 ) -> tuple[dict[str, Any], int]:
@@ -538,76 +233,12 @@ class PromptSizer:
         return content, tok
 
 
-def fetch_and_filter_placements(
-    client: ExoClient, full_model_id: str, args: argparse.Namespace
-) -> list[dict[str, Any]]:
-    previews_resp = client.request_json(
-        "GET", "/instance/previews", params={"model_id": full_model_id}
-    )
-    previews = previews_resp.get("previews") or []
-
-    selected: list[dict[str, Any]] = []
-    for p in previews:
-        if p.get("error") is not None:
-            continue
-        if not placement_filter(str(p.get("instance_meta", "")), args.instance_meta):
-            continue
-        if not sharding_filter(str(p.get("sharding", "")), args.sharding):
-            continue
-
-        instance = p.get("instance")
-        if not isinstance(instance, dict):
-            continue
-
-        n = nodes_used_in_instance(instance)
-        # Skip tensor ring single node as it is pointless when pipeline ring
-        if n == 1 and (
-            (args.sharding == "both" and "tensor" in p.get("sharding", "").lower())
-            or (
-                args.instance_meta == "both"
-                and "jaccl" in p.get("instance_meta", "").lower()
-            )
-        ):
-            continue
-
-        if (
-            args.skip_pipeline_jaccl
-            and (
-                args.instance_meta == "both"
-                and "jaccl" in p.get("instance_meta", "").lower()
-            )
-            and (
-                args.sharding == "both" and "pipeline" in p.get("sharding", "").lower()
-            )
-        ):
-            continue
-
-        if (
-            args.skip_tensor_ring
-            and (
-                args.instance_meta == "both"
-                and "ring" in p.get("instance_meta", "").lower()
-            )
-            and (args.sharding == "both" and "tensor" in p.get("sharding", "").lower())
-        ):
-            continue
-
-        if args.min_nodes <= n <= args.max_nodes:
-            selected.append(p)
-
-    return selected
-
-
 def main() -> int:
     ap = argparse.ArgumentParser(
         prog="exo-bench",
         description="Benchmark exo model throughput across placement previews.",
     )
-    ap.add_argument("--host", default=os.environ.get("EXO_HOST", "localhost"))
-    ap.add_argument(
-        "--port", type=int, default=int(os.environ.get("EXO_PORT", "52415"))
-    )
-    ap.add_argument("--model", required=True, help="Model short id or huggingface id")
+    add_common_instance_args(ap)
     ap.add_argument(
         "--pp",
         nargs="+",
@@ -620,46 +251,21 @@ def main() -> int:
         required=True,
         help="Generation lengths (ints). Accepts commas.",
     )
-    ap.add_argument(
-        "--max-nodes",
-        type=int,
-        default=4,
-        help="Only consider placements using <= this many nodes.",
-    )
-    ap.add_argument(
-        "--min-nodes",
-        type=int,
-        default=1,
-        help="Only consider placements using >= this many nodes.",
-    )
-    ap.add_argument(
-        "--instance-meta", choices=["ring", "jaccl", "both"], default="both"
-    )
-    ap.add_argument(
-        "--sharding", choices=["pipeline", "tensor", "both"], default="both"
-    )
-    ap.add_argument(
-        "--skip-pipeline-jaccl",
-        action="store_true",
-        help="Skip pipeline+jaccl placements, as it's often pointless.",
-    )
-    ap.add_argument(
-        "--skip-tensor-ring",
-        action="store_true",
-        help="Skip tensor+ring placements, as it's so slow.",
-    )
     ap.add_argument(
         "--repeat", type=int, default=1, help="Repetitions per (pp,tg) pair."
     )
+    ap.add_argument(
+        "--concurrency",
+        nargs="+",
+        default=["1"],
+        help="Concurrency levels (ints). Accepts commas. E.g. --concurrency 1,2,4,8. Default 1.",
+    )
     ap.add_argument(
         "--warmup",
         type=int,
         default=0,
         help="Warmup runs per placement (uses first pp/tg).",
     )
-    ap.add_argument(
-        "--timeout", type=float, default=7200.0, help="HTTP timeout (seconds)."
-    )
     ap.add_argument(
         "--json-out",
         default="bench/results.json",
@@ -674,17 +280,6 @@ def main() -> int:
         action="store_true",
         help="Force all pp×tg combinations (cartesian product) even when lists have equal length.",
     )
-    ap.add_argument(
-        "--settle-timeout",
-        type=float,
-        default=0,
-        help="Max seconds to wait for the cluster to produce valid placements (0 = try once).",
-    )
-    ap.add_argument(
-        "--danger-delete-downloads",
-        action="store_true",
-        help="Delete existing models from smallest to largest to make room for benchmark model.",
-    )
     args = ap.parse_args()
 
     pp_list = parse_int_list(args.pp)
@@ -695,6 +290,10 @@ def main() -> int:
     if args.repeat <= 0:
         logger.error("--repeat must be >= 1")
         return 2
+    concurrency_list = parse_int_list(args.concurrency)
+    if not concurrency_list or any(c <= 0 for c in concurrency_list):
+        logger.error("--concurrency values must be >= 1")
+        return 2
 
     # Log pairing mode
     use_combinations = args.all_combinations or len(pp_list) != len(tg_list)
@@ -706,7 +305,9 @@ def main() -> int:
         logger.info(f"pp/tg mode: tandem (zip) - {len(pp_list)} pairs")
 
     client = ExoClient(args.host, args.port, timeout_s=args.timeout)
-    short_id, full_model_id = resolve_model_short_id(client, args.model)
+    short_id, full_model_id = resolve_model_short_id(
+        client, args.model, force_download=args.force_download
+    )
 
     tokenizer = load_tokenizer_for_bench(full_model_id)
     if tokenizer is None:
@@ -719,24 +320,10 @@ def main() -> int:
         logger.error("[exo-bench] tokenizer usable but prompt sizing failed")
         raise
 
-    settle_deadline = (
-        time.monotonic() + args.settle_timeout if args.settle_timeout > 0 else None
+    selected = settle_and_fetch_placements(
+        client, full_model_id, args, settle_timeout=args.settle_timeout
     )
 
-    selected = fetch_and_filter_placements(client, full_model_id, args)
-
-    if not selected and settle_deadline:
-        backoff = _SETTLE_INITIAL_BACKOFF_S
-        while not selected and time.monotonic() < settle_deadline:
-            remaining = settle_deadline - time.monotonic()
-            logger.warning(
-                f"No valid placements yet (cluster may still be settling). "
-                f"Retrying in {backoff:.1f}s ({remaining:.0f}s remaining)..."
-            )
-            time.sleep(min(backoff, remaining))
-            backoff = min(backoff * _SETTLE_BACKOFF_MULTIPLIER, _SETTLE_MAX_BACKOFF_S)
-            selected = fetch_and_filter_placements(client, full_model_id, args)
-
     if not selected:
         logger.error("No valid placements matched your filters.")
         return 1
@@ -760,8 +347,12 @@ def main() -> int:
     if args.dry_run:
         return 0
 
+    settle_deadline = (
+        time.monotonic() + args.settle_timeout if args.settle_timeout > 0 else None
+    )
+
     logger.info("Planning phase: checking downloads...")
-    run_planning_phase(
+    download_duration_s = run_planning_phase(
         client,
         full_model_id,
         selected[0],
@@ -769,6 +360,10 @@ def main() -> int:
         args.timeout,
         settle_deadline,
     )
+    if download_duration_s is not None:
+        logger.info(f"Download: {download_duration_s:.1f}s (freshly downloaded)")
+    else:
+        logger.info("Download: model already cached")
 
     all_rows: list[dict[str, Any]] = []
 
@@ -811,47 +406,155 @@ def main() -> int:
                 pp_tg_pairs = list(zip(pp_list, tg_list, strict=True))
 
             for pp, tg in pp_tg_pairs:
-                runs: list[dict[str, Any]] = []
-                for r in range(args.repeat):
-                    time.sleep(3)
-                    try:
-                        row, actual_pp_tokens = run_one_completion(
-                            client, full_model_id, pp, tg, prompt_sizer
+                for concurrency in concurrency_list:
+                    logger.info(f"--- pp={pp} tg={tg} concurrency={concurrency} ---")
+                    runs: list[dict[str, Any]] = []
+                    for r in range(args.repeat):
+                        time.sleep(3)
+
+                        if concurrency <= 1:
+                            # Sequential: single request
+                            try:
+                                row, actual_pp_tokens = run_one_completion(
+                                    client, full_model_id, pp, tg, prompt_sizer
+                                )
+                            except Exception as e:
+                                logger.error(e)
+                                continue
+                            row.update(
+                                {
+                                    "model_short_id": short_id,
+                                    "model_id": full_model_id,
+                                    "placement_sharding": sharding,
+                                    "placement_instance_meta": instance_meta,
+                                    "placement_nodes": n_nodes,
+                                    "instance_id": instance_id,
+                                    "pp_tokens": actual_pp_tokens,
+                                    "tg": tg,
+                                    "repeat_index": r,
+                                    "concurrency": 1,
+                                    **(
+                                        {"download_duration_s": download_duration_s}
+                                        if download_duration_s is not None
+                                        else {}
+                                    ),
+                                }
+                            )
+                            runs.append(row)
+                            all_rows.append(row)
+                        else:
+                            # Concurrent: fire N requests in parallel
+                            # Pre-build prompt once, barrier ensures simultaneous dispatch
+                            content, actual_pp = prompt_sizer.build(pp)
+                            pre_built_payload: dict[str, Any] = {
+                                "model": full_model_id,
+                                "messages": [{"role": "user", "content": content}],
+                                "stream": False,
+                                "max_tokens": tg,
+                            }
+                            barrier = threading.Barrier(concurrency)
+                            batch_start = threading.Event()
+                            batch_t0: float = 0.0
+                            batch_results: list[tuple[dict[str, Any], int]] = []
+                            batch_errors = 0
+
+                            def _run_concurrent(
+                                idx: int,
+                            ) -> tuple[dict[str, Any], int]:
+                                nonlocal batch_t0
+                                c = ExoClient(
+                                    args.host, args.port, timeout_s=args.timeout
+                                )
+                                if barrier.wait() == 0:
+                                    batch_t0 = time.perf_counter()
+                                    batch_start.set()
+                                else:
+                                    batch_start.wait()
+                                t0 = batch_t0
+                                out = c.post_bench_chat_completions(pre_built_payload)
+                                elapsed = time.perf_counter() - t0
+                                stats = out.get("generation_stats")
+                                choices = out.get("choices") or [{}]
+                                message = choices[0].get("message", {}) if choices else {}
+                                text = message.get("content") or ""
+                                return {
+                                    "elapsed_s": elapsed,
+                                    "output_text_preview": text[:200],
+                                    "stats": stats,
+                                }, actual_pp
+
+                            with ThreadPoolExecutor(max_workers=concurrency) as pool:
+                                futures = {
+                                    pool.submit(_run_concurrent, i): i
+                                    for i in range(concurrency)
+                                }
+                                for fut in as_completed(futures):
+                                    try:
+                                        batch_results.append(fut.result())
+                                    except Exception as e:
+                                        logger.error(f"Concurrent request failed: {e}")
+                                        batch_errors += 1
+                            batch_wall_s = max(x["elapsed_s"] for x, _ in batch_results) if batch_results else time.perf_counter() - batch_t0
+
+                            for idx, (row, actual_pp_tokens) in enumerate(
+                                batch_results
+                            ):
+                                row.update(
+                                    {
+                                        "model_short_id": short_id,
+                                        "model_id": full_model_id,
+                                        "placement_sharding": sharding,
+                                        "placement_instance_meta": instance_meta,
+                                        "placement_nodes": n_nodes,
+                                        "instance_id": instance_id,
+                                        "pp_tokens": actual_pp_tokens,
+                                        "tg": tg,
+                                        "repeat_index": r,
+                                        "concurrency": concurrency,
+                                        "concurrent_index": idx,
+                                        **(
+                                            {"download_duration_s": download_duration_s}
+                                            if download_duration_s is not None
+                                            else {}
+                                        ),
+                                    }
+                                )
+                                runs.append(row)
+                                all_rows.append(row)
+
+                            if batch_results:
+                                valid_gen_tps = [
+                                    x["stats"]["generation_tps"]
+                                    for x, _ in batch_results
+                                    if x["stats"]["generation_tps"] > 0
+                                ]
+                                per_req_tps = max(valid_gen_tps) if valid_gen_tps else 0.0
+                                agg_gen_tps = per_req_tps * concurrency
+                                logger.info(
+                                    f"[concurrent {concurrency}x]  "
+                                    f"agg_gen_tps={agg_gen_tps:.2f}  "
+                                    f"per_req_tps={per_req_tps:.2f}  "
+                                    f"wall_s={batch_wall_s:.2f}  "
+                                    f"errors={batch_errors}"
+                                )
+
+                    if runs:
+                        prompt_tps = mean(x["stats"]["prompt_tps"] for x in runs)
+                        valid_gen = [x["stats"]["generation_tps"] for x in runs if x["stats"]["generation_tps"] > 0]
+                        per_req_tps = max(valid_gen) if valid_gen else 0.0
+                        gen_tps = per_req_tps * concurrency
+                        ptok = mean(x["stats"]["prompt_tokens"] for x in runs)
+                        gtok = mean(x["stats"]["generation_tokens"] for x in runs)
+                        peak = mean(
+                            x["stats"]["peak_memory_usage"]["inBytes"] for x in runs
                         )
-                    except Exception as e:
-                        logger.error(e)
-                        continue
-                    row.update(
-                        {
-                            "model_short_id": short_id,
-                            "model_id": full_model_id,
-                            "placement_sharding": sharding,
-                            "placement_instance_meta": instance_meta,
-                            "placement_nodes": n_nodes,
-                            "instance_id": instance_id,
-                            "pp_tokens": actual_pp_tokens,
-                            "tg": tg,
-                            "repeat_index": r,
-                        }
-                    )
-                    runs.append(row)
-                    all_rows.append(row)
 
-                if runs:
-                    prompt_tps = mean(x["stats"]["prompt_tps"] for x in runs)
-                    gen_tps = mean(x["stats"]["generation_tps"] for x in runs)
-                    ptok = mean(x["stats"]["prompt_tokens"] for x in runs)
-                    gtok = mean(x["stats"]["generation_tokens"] for x in runs)
-                    peak = mean(
-                        x["stats"]["peak_memory_usage"]["inBytes"] for x in runs
-                    )
-
-                    logger.info(
-                        f"prompt_tps={prompt_tps:.2f} gen_tps={gen_tps:.2f}    "
-                        f"prompt_tokens={ptok} gen_tokens={gtok}    "
-                        f"peak_memory={format_peak_memory(peak)}\n"
-                    )
-                time.sleep(2)
+                        logger.info(
+                            f"prompt_tps={prompt_tps:.2f} gen_tps={gen_tps:.2f}    "
+                            f"prompt_tokens={ptok} gen_tokens={gtok}    "
+                            f"peak_memory={format_peak_memory(peak)}\n"
+                        )
+                    time.sleep(2)
         finally:
             try:
                 client.request_json("DELETE", f"/instance/{instance_id}")

bench/harness.py

@@ -0,0 +1,506 @@
+# type: ignore
+from __future__ import annotations
+
+import argparse
+import http.client
+import json
+import os
+import time
+from typing import Any
+from urllib.parse import urlencode
+
+from loguru import logger
+
+_SETTLE_INITIAL_BACKOFF_S = 1.0
+_SETTLE_MAX_BACKOFF_S = 60.0
+_SETTLE_BACKOFF_MULTIPLIER = 2.0
+
+
+class ExoHttpError(RuntimeError):
+    def __init__(self, status: int, reason: str, body_preview: str):
+        super().__init__(f"HTTP {status} {reason}: {body_preview}")
+        self.status = status
+
+
+class ExoClient:
+    def __init__(self, host: str, port: int, timeout_s: float = 7200.0):
+        self.host = host
+        self.port = port
+        self.timeout_s = timeout_s
+
+    def request_json(
+        self,
+        method: str,
+        path: str,
+        params: dict[str, Any] | None = None,
+        body: dict[str, Any] | None = None,
+        headers: dict[str, str] | None = None,
+    ) -> Any:
+        if not path.startswith("/"):
+            path = "/" + path
+        if params:
+            path = path + "?" + urlencode(params)
+
+        conn = http.client.HTTPConnection(self.host, self.port, timeout=self.timeout_s)
+        try:
+            payload: bytes | None = None
+            hdrs: dict[str, str] = {"Accept": "application/json"}
+
+            if body is not None:
+                payload = json.dumps(body).encode("utf-8")
+                hdrs["Content-Type"] = "application/json"
+            if headers:
+                hdrs.update(headers)
+
+            conn.request(method.upper(), path, body=payload, headers=hdrs)
+            resp = conn.getresponse()
+            raw = resp.read()
+            text = raw.decode("utf-8", errors="replace") if raw else ""
+
+            if resp.status >= 400:
+                raise ExoHttpError(resp.status, resp.reason, text[:300])
+
+            if not text:
+                return None
+            return json.loads(text)
+        finally:
+            conn.close()
+
+    def post_bench_chat_completions(self, payload: dict[str, Any]) -> dict[str, Any]:
+        return self.request_json("POST", "/bench/chat/completions", body=payload)
+
+
+def unwrap_instance(instance: dict[str, Any]) -> dict[str, Any]:
+    if len(instance) != 1:
+        raise KeyError(f"Expected 1 key, got keys={list(instance.keys())}")
+
+    tag = next(iter(instance))
+    inner = instance[tag]
+    if not isinstance(inner, dict):
+        raise TypeError(f"payload for {tag} must be dict, got {type(inner)}")
+    return inner
+
+
+def instance_id_from_instance(instance: dict[str, Any]) -> str:
+    inner = unwrap_instance(instance)
+    return str(inner["instanceId"])
+
+
+def nodes_used_in_instance(instance: dict[str, Any]) -> int:
+    inner = unwrap_instance(instance)
+    return len(inner["shardAssignments"]["nodeToRunner"])
+
+
+def runner_ids_from_instance(instance: dict[str, Any]) -> list[str]:
+    inner = unwrap_instance(instance)
+    runner_to_shard = inner["shardAssignments"]["runnerToShard"]
+    return list(runner_to_shard.keys())
+
+
+def runner_ready(runner: dict[str, Any]) -> bool:
+    return "RunnerReady" in runner
+
+
+def runner_failed(runner: dict[str, Any]) -> bool:
+    return "RunnerFailed" in runner
+
+
+def get_runner_failed_message(runner: dict[str, Any]) -> str | None:
+    if "RunnerFailed" in runner:
+        return runner["RunnerFailed"].get("errorMessage")
+    return None
+
+
+def wait_for_instance_ready(
+    client: ExoClient, instance_id: str, timeout: float = 24000.0
+) -> None:
+    start_time = time.time()
+    instance_existed = False
+    while time.time() - start_time < timeout:
+        state = client.request_json("GET", "/state")
+        instances = state.get("instances", {})
+
+        if instance_id not in instances:
+            if instance_existed:
+                # Instance was deleted after being created - likely due to runner failure
+                raise RuntimeError(
+                    f"Instance {instance_id} was deleted (runner may have failed)"
+                )
+            time.sleep(0.1)
+            continue
+
+        instance_existed = True
+        instance = instances[instance_id]
+        runner_ids = runner_ids_from_instance(instance)
+        runners = state.get("runners", {})
+
+        # Check for failed runners first
+        for rid in runner_ids:
+            runner = runners.get(rid, {})
+            if runner_failed(runner):
+                error_msg = get_runner_failed_message(runner) or "Unknown error"
+                raise RuntimeError(f"Runner {rid} failed: {error_msg}")
+
+        if all(runner_ready(runners.get(rid, {})) for rid in runner_ids):
+            return
+
+        time.sleep(0.1)
+
+    raise TimeoutError(f"Instance {instance_id} did not become ready within {timeout=}")
+
+
+def wait_for_instance_gone(
+    client: ExoClient, instance_id: str, timeout: float = 3.0
+) -> None:
+    start_time = time.time()
+    while time.time() - start_time < timeout:
+        try:
+            client.request_json("GET", f"/instance/{instance_id}")
+            time.sleep(0.4)
+        except ExoHttpError as e:
+            if e.status == 404:
+                return
+            raise
+
+    raise TimeoutError(f"Instance {instance_id} did not get deleted within {timeout=}")
+
+
+def resolve_model_short_id(
+    client: ExoClient, model_arg: str, *, force_download: bool = False
+) -> tuple[str, str]:
+    models = client.request_json("GET", "/models") or {}
+    data = models.get("data") or []
+
+    for m in data:
+        if (m.get("name") or "").lower() == model_arg.lower():
+            short_id = str(m["name"])
+            full_id = str(m.get("hugging_face_id") or m["name"])
+            return short_id, full_id
+
+    for m in data:
+        if m.get("hugging_face_id") == model_arg:
+            short_id = str(m["name"])
+            full_id = str(m["hugging_face_id"])
+            return short_id, full_id
+
+    if force_download and "/" in model_arg:
+        logger.info(f"Model not in /models, adding from HuggingFace: {model_arg}")
+        result = client.request_json(
+            "POST", "/models/add", body={"model_id": model_arg}
+        )
+        if result:
+            short_id = str(result.get("name") or model_arg.rsplit("/", 1)[-1])
+            full_id = str(result.get("hugging_face_id") or model_arg)
+            return short_id, full_id
+
+    raise ValueError(f"Model not found in /models: {model_arg}")
+
+
+def placement_filter(instance_meta: str, wanted: str) -> bool:
+    s = (instance_meta or "").lower()
+    if wanted == "both":
+        return ("ring" in s) or ("jaccl" in s)
+    return wanted in s
+
+
+def sharding_filter(sharding: str, wanted: str) -> bool:
+    s = (sharding or "").lower()
+    if wanted == "both":
+        return ("pipeline" in s) or ("tensor" in s)
+    return wanted in s
+
+
+def fetch_and_filter_placements(
+    client: ExoClient, full_model_id: str, args: argparse.Namespace
+) -> list[dict[str, Any]]:
+    previews_resp = client.request_json(
+        "GET", "/instance/previews", params={"model_id": full_model_id}
+    )
+    previews = previews_resp.get("previews") or []
+
+    selected: list[dict[str, Any]] = []
+    for p in previews:
+        if p.get("error") is not None:
+            continue
+        if not placement_filter(str(p.get("instance_meta", "")), args.instance_meta):
+            continue
+        if not sharding_filter(str(p.get("sharding", "")), args.sharding):
+            continue
+
+        instance = p.get("instance")
+        if not isinstance(instance, dict):
+            continue
+
+        n = nodes_used_in_instance(instance)
+        # Skip tensor ring single node as it is pointless when pipeline ring
+        if n == 1 and (
+            (args.sharding == "both" and "tensor" in p.get("sharding", "").lower())
+            or (
+                args.instance_meta == "both"
+                and "jaccl" in p.get("instance_meta", "").lower()
+            )
+        ):
+            continue
+
+        if (
+            args.skip_pipeline_jaccl
+            and (
+                args.instance_meta == "both"
+                and "jaccl" in p.get("instance_meta", "").lower()
+            )
+            and (
+                args.sharding == "both" and "pipeline" in p.get("sharding", "").lower()
+            )
+        ):
+            continue
+
+        if (
+            args.skip_tensor_ring
+            and (
+                args.instance_meta == "both"
+                and "ring" in p.get("instance_meta", "").lower()
+            )
+            and (args.sharding == "both" and "tensor" in p.get("sharding", "").lower())
+        ):
+            continue
+
+        if args.min_nodes <= n <= args.max_nodes:
+            selected.append(p)
+
+    return selected
+
+
+def settle_and_fetch_placements(
+    client: ExoClient,
+    full_model_id: str,
+    args: argparse.Namespace,
+    settle_timeout: float = 0,
+) -> list[dict[str, Any]]:
+    selected = fetch_and_filter_placements(client, full_model_id, args)
+
+    if not selected and settle_timeout > 0:
+        backoff = _SETTLE_INITIAL_BACKOFF_S
+        deadline = time.monotonic() + settle_timeout
+        while not selected and time.monotonic() < deadline:
+            remaining = deadline - time.monotonic()
+            logger.warning(
+                f"No valid placements yet (cluster may still be settling). "
+                f"Retrying in {backoff:.1f}s ({remaining:.0f}s remaining)..."
+            )
+            time.sleep(min(backoff, remaining))
+            backoff = min(backoff * _SETTLE_BACKOFF_MULTIPLIER, _SETTLE_MAX_BACKOFF_S)
+            selected = fetch_and_filter_placements(client, full_model_id, args)
+
+    return selected
+
+
+def run_planning_phase(
+    client: ExoClient,
+    full_model_id: str,
+    preview: dict[str, Any],
+    danger_delete: bool,
+    timeout: float,
+    settle_deadline: float | None,
+) -> float | None:
+    """Check disk space and ensure model is downloaded before benchmarking.
+
+    Returns the wall-clock download duration in seconds if a fresh download
+    was needed, or None if the model was already cached on all nodes.
+    """
+    # Get model size from /models
+    models = client.request_json("GET", "/models") or {}
+    model_bytes = 0
+    for m in models.get("data", []):
+        if m.get("hugging_face_id") == full_model_id:
+            model_bytes = m.get("storage_size_megabytes", 0) * 1024 * 1024
+            break
+
+    if not model_bytes:
+        logger.warning(
+            f"Could not determine size for {full_model_id}, skipping disk check"
+        )
+        return None
+
+    # Get nodes from preview
+    inner = unwrap_instance(preview["instance"])
+    node_ids = list(inner["shardAssignments"]["nodeToRunner"].keys())
+    runner_to_shard = inner["shardAssignments"]["runnerToShard"]
+
+    state = client.request_json("GET", "/state")
+    downloads = state.get("downloads", {})
+    node_disk = state.get("nodeDisk", {})
+
+    needs_download = False
+
+    for node_id in node_ids:
+        node_downloads = downloads.get(node_id, [])
+
+        # Check if model already downloaded on this node
+        already_downloaded = any(
+            "DownloadCompleted" in p
+            and unwrap_instance(p["DownloadCompleted"]["shardMetadata"])["modelCard"][
+                "modelId"
+            ]
+            == full_model_id
+            for p in node_downloads
+        )
+        if already_downloaded:
+            continue
+
+        needs_download = True
+
+        # Wait for disk info if settle_deadline is set
+        disk_info = node_disk.get(node_id, {})
+        backoff = _SETTLE_INITIAL_BACKOFF_S
+        while not disk_info and settle_deadline and time.monotonic() < settle_deadline:
+            remaining = settle_deadline - time.monotonic()
+            logger.info(
+                f"Waiting for disk info on {node_id} ({remaining:.0f}s remaining)..."
+            )
+            time.sleep(min(backoff, remaining))
+            backoff = min(backoff * _SETTLE_BACKOFF_MULTIPLIER, _SETTLE_MAX_BACKOFF_S)
+            state = client.request_json("GET", "/state")
+            node_disk = state.get("nodeDisk", {})
+            disk_info = node_disk.get(node_id, {})
+
+        if not disk_info:
+            logger.warning(f"No disk info for {node_id}, skipping space check")
+            continue
+
+        avail = disk_info.get("available", {}).get("inBytes", 0)
+        if avail >= model_bytes:
+            continue
+
+        if not danger_delete:
+            raise RuntimeError(
+                f"Insufficient disk on {node_id}: need {model_bytes // (1024**3)}GB, "
+                f"have {avail // (1024**3)}GB. Use --danger-delete-downloads to free space."
+            )
+
+        # Delete from smallest to largest (skip read-only models)
+        completed = [
+            (
+                unwrap_instance(p["DownloadCompleted"]["shardMetadata"])["modelCard"][
+                    "modelId"
+                ],
+                p["DownloadCompleted"]["total"]["inBytes"],
+            )
+            for p in node_downloads
+            if "DownloadCompleted" in p
+            and not p["DownloadCompleted"].get("readOnly", False)
+        ]
+        for del_model, size in sorted(completed, key=lambda x: x[1]):
+            logger.info(f"Deleting {del_model} from {node_id} ({size // (1024**2)}MB)")
+            client.request_json("DELETE", f"/download/{node_id}/{del_model}")
+            avail += size
+            if avail >= model_bytes:
+                break
+
+        if avail < model_bytes:
+            raise RuntimeError(f"Could not free enough space on {node_id}")
+
+    # Start downloads (idempotent)
+    download_t0 = time.perf_counter() if needs_download else None
+    for node_id in node_ids:
+        runner_id = inner["shardAssignments"]["nodeToRunner"][node_id]
+        shard = runner_to_shard[runner_id]
+        client.request_json(
+            "POST",
+            "/download/start",
+            body={
+                "targetNodeId": node_id,
+                "shardMetadata": shard,
+            },
+        )
+        logger.info(f"Started download on {node_id}")
+
+    # Wait for downloads
+    start = time.time()
+    while time.time() - start < timeout:
+        state = client.request_json("GET", "/state")
+        downloads = state.get("downloads", {})
+        all_done = True
+        for node_id in node_ids:
+            done = any(
+                "DownloadCompleted" in p
+                and unwrap_instance(p["DownloadCompleted"]["shardMetadata"])[
+                    "modelCard"
+                ]["modelId"]
+                == full_model_id
+                for p in downloads.get(node_id, [])
+            )
+            failed = [
+                p["DownloadFailed"]["errorMessage"]
+                for p in downloads.get(node_id, [])
+                if "DownloadFailed" in p
+                and unwrap_instance(p["DownloadFailed"]["shardMetadata"])["modelCard"][
+                    "modelId"
+                ]
+                == full_model_id
+            ]
+            if failed:
+                raise RuntimeError(f"Download failed on {node_id}: {failed[0]}")
+            if not done:
+                all_done = False
+        if all_done:
+            if download_t0 is not None:
+                return time.perf_counter() - download_t0
+            return None
+        time.sleep(1)
+
+    raise TimeoutError("Downloads did not complete in time")
+
+
+def add_common_instance_args(ap: argparse.ArgumentParser) -> None:
+    ap.add_argument("--host", default=os.environ.get("EXO_HOST", "localhost"))
+    ap.add_argument(
+        "--port", type=int, default=int(os.environ.get("EXO_PORT", "52415"))
+    )
+    ap.add_argument("--model", required=True, help="Model short id or huggingface id")
+    ap.add_argument(
+        "--force-download",
+        action="store_true",
+        help="If model not in /models, add it from HuggingFace via exo and download.",
+    )
+    ap.add_argument(
+        "--max-nodes",
+        type=int,
+        default=4,
+        help="Only consider placements using <= this many nodes.",
+    )
+    ap.add_argument(
+        "--min-nodes",
+        type=int,
+        default=1,
+        help="Only consider placements using >= this many nodes.",
+    )
+    ap.add_argument(
+        "--instance-meta", choices=["ring", "jaccl", "both"], default="both"
+    )
+    ap.add_argument(
+        "--sharding", choices=["pipeline", "tensor", "both"], default="both"
+    )
+    ap.add_argument(
+        "--skip-pipeline-jaccl",
+        action="store_true",
+        help="Skip pipeline+jaccl placements, as it's often pointless.",
+    )
+    ap.add_argument(
+        "--skip-tensor-ring",
+        action="store_true",
+        help="Skip tensor+ring placements, as it's so slow.",
+    )
+    ap.add_argument(
+        "--timeout", type=float, default=7200.0, help="HTTP timeout (seconds)."
+    )
+    ap.add_argument(
+        "--settle-timeout",
+        type=float,
+        default=0,
+        help="Max seconds to wait for the cluster to produce valid placements (0 = try once).",
+    )
+    ap.add_argument(
+        "--danger-delete-downloads",
+        action="store_true",
+        help="Delete existing models from smallest to largest to make room for benchmark model.",
+    )

bench/parallel_requests.py

@@ -0,0 +1,255 @@
+# type: ignore
+import argparse
+import asyncio
+import sys
+import termios
+import time
+import tty
+
+import aiohttp
+
+NUM_REQUESTS = 10
+BASE_URL = ""
+
+QUESTIONS = [
+    "What is the capital of Australia?",
+    "How many bones are in the human body?",
+    "What year did World War II end?",
+    "What is the speed of light in meters per second?",
+    "Who wrote Romeo and Juliet?",
+    "What is the chemical formula for water?",
+    "How many planets are in our solar system?",
+    "What is the largest ocean on Earth?",
+    "Who painted the Mona Lisa?",
+    "What is the boiling point of water in Celsius?",
+]
+
+
+def write(s: str) -> None:
+    sys.stdout.write(s)
+
+
+# ---------------------------------------------------------------------------
+# Model picker (same style as exo_eval)
+# ---------------------------------------------------------------------------
+
+
+def fetch_models() -> list[str]:
+    import json
+    import urllib.request
+
+    with urllib.request.urlopen(f"{BASE_URL}/state") as resp:
+        data = json.loads(resp.read())
+    model_ids: set[str] = set()
+    for instance in data.get("instances", {}).values():
+        for variant in instance.values():
+            sa = variant.get("shardAssignments", {})
+            model_id = sa.get("modelId")
+            if model_id:
+                model_ids.add(model_id)
+    return sorted(model_ids)
+
+
+def pick_model() -> str | None:
+    models = fetch_models()
+    if not models:
+        print("No models found.")
+        return None
+
+    cursor = 0
+    total_lines = len(models) + 4
+
+    def render(first: bool = False) -> None:
+        if not first:
+            write(f"\033[{total_lines}A")
+        write("\033[J")
+        write("\033[1mSelect model\033[0m (up/down, enter confirm, q quit)\r\n\r\n")
+        for i, model in enumerate(models):
+            line = f"  {'>' if i == cursor else ' '} {model}"
+            write(f"\033[7m{line}\033[0m\r\n" if i == cursor else f"{line}\r\n")
+        write("\r\n")
+        sys.stdout.flush()
+
+    fd = sys.stdin.fileno()
+    old = termios.tcgetattr(fd)
+    try:
+        tty.setraw(fd)
+        write("\033[?25l")
+        render(first=True)
+        while True:
+            ch = sys.stdin.read(1)
+            if ch in ("q", "\x03"):
+                write("\033[?25h\033[0m\r\n")
+                return None
+            elif ch in ("\r", "\n"):
+                break
+            elif ch == "\x1b":
+                seq = sys.stdin.read(2)
+                if seq == "[A":
+                    cursor = (cursor - 1) % len(models)
+                elif seq == "[B":
+                    cursor = (cursor + 1) % len(models)
+            render()
+    finally:
+        termios.tcsetattr(fd, termios.TCSADRAIN, old)
+        write(f"\033[{total_lines}A\033[J")  # clear picker UI
+        write("\033[?25h\033[0m")
+        sys.stdout.flush()
+
+    return models[cursor]
+
+
+# ---------------------------------------------------------------------------
+# Parallel requests
+# ---------------------------------------------------------------------------
+
+statuses: list[str] = []
+times: list[str] = []
+previews: list[str] = []
+tokens: list[str] = []
+full_responses: list[dict | None] = []
+total_lines = 0
+start_time: float = 0
+selected_model: str = ""
+
+
+def render_progress(first: bool = False) -> None:
+    if not first:
+        write(f"\033[{total_lines}A")
+    write("\033[J")
+    elapsed = time.monotonic() - start_time if start_time else 0
+    done = sum(1 for s in statuses if s == "done")
+    write(
+        f"\033[1m{selected_model}\033[0m  [{done}/{NUM_REQUESTS}]  {elapsed:.1f}s\r\n\r\n"
+    )
+
+    for i in range(NUM_REQUESTS):
+        q = QUESTIONS[i % len(QUESTIONS)]
+        status = statuses[i]
+        if status == "pending":
+            color = "\033[33m"  # yellow
+        elif status == "running":
+            color = "\033[36m"  # cyan
+        elif status == "done":
+            color = "\033[32m"  # green
+        else:
+            color = "\033[31m"  # red
+        write(
+            f"  {i:>2}  {color}{status:<8}\033[0m  {times[i]:>6}  {tokens[i]:>5}tok  {q[:40]:<40}  {previews[i][:50]}\r\n"
+        )
+
+    write("\r\n")
+    sys.stdout.flush()
+
+
+async def send_request(
+    session: aiohttp.ClientSession, i: int, lock: asyncio.Lock
+) -> None:
+    payload = {
+        "model": selected_model,
+        "messages": [{"role": "user", "content": QUESTIONS[i % len(QUESTIONS)]}],
+        "max_tokens": 1024,
+    }
+    statuses[i] = "running"
+    async with lock:
+        render_progress()
+    t0 = time.monotonic()
+    try:
+        async with session.post(
+            f"{BASE_URL}/v1/chat/completions", json=payload
+        ) as resp:
+            data = await resp.json()
+            elapsed = time.monotonic() - t0
+            full_responses[i] = data
+            times[i] = f"{elapsed:.1f}s"
+            if resp.status == 200:
+                choice = data["choices"][0]
+                msg = choice["message"]
+                content = msg.get("content", "")
+                previews[i] = content[:50].replace("\n", " ") or "(empty)"
+                if "usage" in data:
+                    tokens[i] = str(data["usage"].get("total_tokens", ""))
+                statuses[i] = "done"
+            else:
+                statuses[i] = f"err:{resp.status}"
+                previews[i] = str(data.get("error", {}).get("message", ""))[:50]
+    except Exception as e:
+        elapsed = time.monotonic() - t0
+        times[i] = f"{elapsed:.1f}s"
+        statuses[i] = "error"
+        previews[i] = str(e)[:50]
+    async with lock:
+        render_progress()
+
+
+async def run_requests(print_stdout: bool = False) -> None:
+    global start_time, total_lines, statuses, times, previews, tokens, full_responses
+
+    statuses = ["pending"] * NUM_REQUESTS
+    times = ["-"] * NUM_REQUESTS
+    previews = ["-"] * NUM_REQUESTS
+    tokens = ["-"] * NUM_REQUESTS
+    full_responses = [None] * NUM_REQUESTS
+    total_lines = NUM_REQUESTS + 4
+
+    write("\033[?25l")  # hide cursor
+    start_time = time.monotonic()
+    render_progress(first=True)
+    lock = asyncio.Lock()
+    try:
+        async with aiohttp.ClientSession() as session:
+            tasks = [send_request(session, i, lock) for i in range(NUM_REQUESTS)]
+            await asyncio.gather(*tasks)
+        total = time.monotonic() - start_time
+        write(
+            f"\033[1m=== All {NUM_REQUESTS} requests done in {total:.1f}s ===\033[0m\r\n\r\n"
+        )
+
+        if print_stdout:
+            for i in range(NUM_REQUESTS):
+                data = full_responses[i]
+                if not data or "choices" not in data:
+                    continue
+                choice = data["choices"][0]
+                msg = choice["message"]
+                q = QUESTIONS[i % len(QUESTIONS)]
+                write(f"\033[1m--- #{i}: {q} ---\033[0m\r\n")
+                if msg.get("reasoning_content"):
+                    write(f"\033[2m[Thinking]: {msg['reasoning_content']}\033[0m\r\n")
+                write(f"{msg.get('content', '')}\r\n")
+                if "usage" in data:
+                    u = data["usage"]
+                    write(
+                        f"\033[2m[Usage: prompt={u.get('prompt_tokens')}, "
+                        f"completion={u.get('completion_tokens')}, "
+                        f"total={u.get('total_tokens')}]\033[0m\r\n"
+                    )
+                write("\r\n")
+    finally:
+        write("\033[?25h")  # show cursor
+        sys.stdout.flush()
+
+
+def main() -> None:
+    global selected_model, BASE_URL
+    parser = argparse.ArgumentParser(
+        description="Send parallel requests to an exo cluster"
+    )
+    parser.add_argument(
+        "--host", required=True, help="Hostname of the exo node (e.g. s1)"
+    )
+    parser.add_argument("--port", type=int, default=52415, help="Port (default: 52415)")
+    parser.add_argument(
+        "--stdout", action="store_true", help="Print full responses after completion"
+    )
+    args = parser.parse_args()
+    BASE_URL = f"http://{args.host}:{args.port}"
+    model = pick_model()
+    if not model:
+        return
+    selected_model = model
+    asyncio.run(run_requests(print_stdout=args.stdout))
+
+
+if __name__ == "__main__":
+    main()

bench/pyproject.toml

@@ -4,11 +4,18 @@ version = "0.1.0"
 description = "Benchmarking tool for exo distributed inference"
 requires-python = ">=3.13"
 dependencies = [
-    "loguru>=0.7.3",
-    "transformers>=5.0.0",
-    "huggingface-hub>=0.33.4",
-    "tiktoken>=0.12.0",
-    "jinja2>=3.1.0",
+  "httpx>=0.27.0",
+  "loguru>=0.7.3",
+  "transformers>=5.0.0",
+  "huggingface-hub>=0.33.4",
+  "tiktoken>=0.12.0",
+  "jinja2>=3.1.0",
+  "protobuf>=5.29.0",
+  "datasets>=2.0.0",
+  "math-verify>=0.7.0",
+  "lm-eval[api,math]>=0.4.0",
+  "human-eval>=1.0.3",
+  "numpy>=1.24.0",
 ]
 
 [build-system]

bench/scenarios.toml

@@ -0,0 +1,306 @@
+# Tool definitions — each becomes an OpenAI function tool.
+# All scenarios get all tools unless they specify a `tools` list.
+
+[tools.get_current_weather]
+description = "Get the current weather in a given location"
+required = ["location"]
+
+[tools.get_current_weather.properties.location]
+type = "string"
+description = "City and state, e.g. San Francisco, CA"
+
+[tools.get_current_weather.properties.unit]
+type = "string"
+enum = ["celsius", "fahrenheit"]
+description = "Temperature unit"
+
+[tools.calculate]
+description = "Evaluate a mathematical expression and return the numeric result"
+required = ["expression"]
+
+[tools.calculate.properties.expression]
+type = "string"
+description = "The math expression to evaluate, e.g. '2 + 3 * 4'"
+
+[tools.search_products]
+description = "Search for products in a catalog by query, category, and price"
+required = ["query"]
+
+[tools.search_products.properties.query]
+type = "string"
+description = "Search query string"
+
+[tools.search_products.properties.category]
+type = "string"
+enum = ["electronics", "clothing", "food", "books"]
+description = "Product category to filter by"
+
+[tools.search_products.properties.max_price]
+type = "number"
+description = "Maximum price in USD"
+
+[tools.create_todos]
+description = "Create a structured todo list"
+required = ["todos"]
+
+[tools.create_todos.properties.todos]
+type = "array"
+description = "List of todo items"
+
+[tools.create_todos.properties.todos.items]
+type = "object"
+required = ["content", "status", "priority"]
+
+[tools.create_todos.properties.todos.items.properties.content]
+type = "string"
+description = "The todo item text"
+
+[tools.create_todos.properties.todos.items.properties.status]
+type = "string"
+description = "Status: pending, in_progress, or completed"
+
+[tools.create_todos.properties.todos.items.properties.priority]
+type = "string"
+description = "Priority: low, normal, or high"
+
+# -- Should call a tool --
+
+[[scenarios]]
+name = "weather_simple"
+description = "Basic weather query -> get_current_weather"
+expect_tool_call = true
+expected_function = "get_current_weather"
+required_arg_keys = ["location"]
+
+[[scenarios.messages]]
+role = "user"
+content = "What's the weather like in Tokyo right now?"
+
+[[scenarios]]
+name = "calculator_simple"
+description = "Math question -> calculate"
+expect_tool_call = true
+expected_function = "calculate"
+required_arg_keys = ["expression"]
+
+[[scenarios.messages]]
+role = "user"
+content = "Use the calculator to compute 3847 * 926 + 17293"
+
+[[scenarios]]
+name = "search_with_filters"
+description = "Product search with category and price filter"
+expect_tool_call = true
+expected_function = "search_products"
+required_arg_keys = ["query"]
+
+[[scenarios.messages]]
+role = "user"
+content = "Find me electronics under $50"
+
+# -- Multi-turn: tool call then follow-up --
+
+[[scenarios]]
+name = "weather_multi_turn"
+description = "Weather query -> tool result -> natural language summary"
+expect_tool_call = true
+expected_function = "get_current_weather"
+required_arg_keys = ["location"]
+
+[scenarios.tool_result]
+temperature = "18C"
+condition = "partly cloudy"
+humidity = "65%"
+wind = "12 km/h NW"
+
+[[scenarios.messages]]
+role = "user"
+content = "What's the weather in Paris?"
+
+[[scenarios]]
+name = "calculator_multi_turn"
+description = "Math query -> tool result -> model reports the answer"
+expect_tool_call = true
+expected_function = "calculate"
+required_arg_keys = ["expression"]
+
+[scenarios.tool_result]
+result = 491682
+
+[[scenarios.messages]]
+role = "user"
+content = "Use the calculator to compute 1847 * 263 + 5921"
+
+[[scenarios]]
+name = "search_multi_turn"
+description = "Search query -> tool result -> model summarizes products"
+expect_tool_call = true
+expected_function = "search_products"
+required_arg_keys = ["query"]
+
+[[scenarios.tool_result.results]]
+name = "Hands-On Machine Learning"
+price = 45.99
+rating = 4.8
+
+[[scenarios.tool_result.results]]
+name = "Deep Learning with Python"
+price = 39.99
+rating = 4.6
+
+[[scenarios.messages]]
+role = "user"
+content = "Search for books about machine learning"
+
+# -- Sequential tool calls --
+
+[[scenarios]]
+name = "chained_tool_calls_same"
+description = "Thinking + weather(Tokyo) -> result -> model must call weather(London)"
+expect_tool_call = true
+expected_function = "get_current_weather"
+required_arg_keys = ["location"]
+
+[[scenarios.messages]]
+role = "user"
+content = "Compare the weather in Tokyo and London."
+
+[[scenarios.messages]]
+role = "assistant"
+content = "I'll check both cities. Let me start with Tokyo."
+
+[[scenarios.messages.tool_calls]]
+id = "call_1"
+name = "get_current_weather"
+arguments = { location = "Tokyo" }
+
+[[scenarios.messages]]
+role = "tool"
+tool_call_id = "call_1"
+content = '{"temperature": "25C", "condition": "sunny"}'
+
+[[scenarios]]
+name = "chained_tool_calls_different"
+description = "Thinking + weather(Berlin) -> result -> model must call calculator"
+expect_tool_call = true
+expected_function = "calculate"
+required_arg_keys = ["expression"]
+
+[[scenarios.messages]]
+role = "user"
+content = "What's the weather in Berlin, and also use the calculator to compute 4819 * 37 + 291."
+
+[[scenarios.messages]]
+role = "assistant"
+content = "I'll handle both. Let me check Berlin's weather first."
+
+[[scenarios.messages.tool_calls]]
+id = "call_2"
+name = "get_current_weather"
+arguments = { location = "Berlin" }
+
+[[scenarios.messages]]
+role = "tool"
+tool_call_id = "call_2"
+content = '{"temperature": "12C", "condition": "rainy"}'
+
+[[scenarios]]
+name = "chained_tool_calls_three"
+description = "Two prior thinking+tool calls -> results -> model must make a third"
+expect_tool_call = true
+expected_function = "get_current_weather"
+required_arg_keys = ["location"]
+
+[[scenarios.messages]]
+role = "user"
+content = "Compare weather in Tokyo, Paris, and London."
+
+[[scenarios.messages]]
+role = "assistant"
+content = "I'll check all three cities. Starting with Tokyo."
+
+[[scenarios.messages.tool_calls]]
+id = "call_3"
+name = "get_current_weather"
+arguments = { location = "Tokyo" }
+
+[[scenarios.messages]]
+role = "tool"
+tool_call_id = "call_3"
+content = '{"temperature": "25C", "condition": "sunny"}'
+
+[[scenarios.messages]]
+role = "assistant"
+content = "Got Tokyo. Now checking Paris."
+
+[[scenarios.messages.tool_calls]]
+id = "call_4"
+name = "get_current_weather"
+arguments = { location = "Paris" }
+
+[[scenarios.messages]]
+role = "tool"
+tool_call_id = "call_4"
+content = '{"temperature": "18C", "condition": "cloudy"}'
+
+# -- Nested object schema (regression for lossy chat template rendering) --
+
+[[scenarios]]
+name = "nested_schema_tool_call"
+description = "Tool call with nested object array schema -> create_todos"
+expect_tool_call = true
+expected_function = "create_todos"
+required_arg_keys = ["todos"]
+nested_array_key = "todos"
+required_item_keys = ["content", "status", "priority"]
+tools = ["create_todos"]
+
+[[scenarios.messages]]
+role = "user"
+content = "Create a todo list with 3 items to learn Python"
+
+# -- Tool name integrity (regression for harmony token leaking into name) --
+
+[tools.glob]
+description = "Search for files matching a glob pattern in the codebase"
+required = ["pattern"]
+
+[tools.glob.properties.pattern]
+type = "string"
+description = "The glob pattern to match files against, e.g. '**/*.py'"
+
+[tools.glob.properties.path]
+type = "string"
+description = "The directory to search in"
+
+[[scenarios]]
+name = "tool_name_integrity"
+description = "Tool name must not contain harmony tokens like <|channel|>"
+expect_tool_call = true
+expected_function = "glob"
+required_arg_keys = ["pattern"]
+tools = ["glob"]
+
+[[scenarios.messages]]
+role = "user"
+content = "Find all Python files in the src directory"
+
+# -- Should NOT call a tool --
+
+[[scenarios]]
+name = "no_tool_joke"
+description = "Joke request should NOT trigger any tool"
+expect_tool_call = false
+
+[[scenarios.messages]]
+role = "user"
+content = "Tell me a funny joke about cats."
+
+[[scenarios]]
+name = "no_tool_factual"
+description = "Factual question answerable from training data"
+expect_tool_call = false
+
+[[scenarios.messages]]
+role = "user"
+content = "What is the capital of Japan?"

bench/single-m3-ultra.toml

@@ -2,10 +2,10 @@
 #
 # Shared constraints applied to ALL benchmarks in this file.
 constraints = [
-    "All(MacOsBuild(=25D125))",
-    "Hosts(=1)",
-    "All(Chip(m3_ultra))",
-    "All(GpuCores(=80))",
+  "All(MacOsBuild(=25D125))",
+  "Hosts(=1)",
+  "All(Chip(m3_ultra))",
+  "All(GpuCores(=80))",
 ]
 
 [topology]

bench/vendor/lcb_testing_util.py

@@ -0,0 +1,593 @@
+# type: ignore
+# Vendored from LiveCodeBench (https://github.com/LiveCodeBench/LiveCodeBench)
+# File: lcb_runner/evaluation/testing_util.py
+# License: MIT
+# Vendored 2026-03-07 — do not modify without updating from upstream.
+
+import ast
+import faulthandler
+import json
+import platform
+
+# to run the solution files we're using a timing based approach
+import signal
+import sys
+import time
+
+# used for debugging to time steps
+from datetime import datetime
+from decimal import Decimal
+from enum import Enum
+from io import StringIO
+
+# from pyext import RuntimeModule
+from types import ModuleType
+
+# used for testing the code that reads from input
+from unittest.mock import mock_open, patch
+
+import_string = "from string import *\nfrom re import *\nfrom datetime import *\nfrom collections import *\nfrom heapq import *\nfrom bisect import *\nfrom copy import *\nfrom math import *\nfrom random import *\nfrom statistics import *\nfrom itertools import *\nfrom functools import *\nfrom operator import *\nfrom io import *\nfrom sys import *\nfrom json import *\nfrom builtins import *\nfrom typing import *\nimport string\nimport re\nimport datetime\nimport collections\nimport heapq\nimport bisect\nimport copy\nimport math\nimport random\nimport statistics\nimport itertools\nimport functools\nimport operator\nimport io\nimport sys\nimport json\nsys.setrecursionlimit(50000)\n"
+
+
+def truncatefn(s, length=300):
+    if isinstance(s, str):
+        pass
+    else:
+        s = str(s)
+    if len(s) <= length:
+        return s
+
+    return s[: length // 2] + "...(truncated) ..." + s[-length // 2 :]
+
+
+class CODE_TYPE(Enum):
+    call_based = 0
+    standard_input = 1
+
+
+# stuff for setting up signal timer
+class TimeoutException(Exception):
+    pass
+
+
+def timeout_handler(signum, frame):
+    print("timeout occured: alarm went off")
+    raise TimeoutException
+
+
+# used to capture stdout as a list
+# from https://stackoverflow.com/a/16571630/6416660
+# alternative use redirect_stdout() from contextlib
+class Capturing(list):
+    def __enter__(self):
+        self._stdout = sys.stdout
+        sys.stdout = self._stringio = StringIO()
+        # Make closing the StringIO a no-op
+        self._stringio.close = lambda x: 1
+        return self
+
+    def __exit__(self, *args):
+        self.append(self._stringio.getvalue())
+        del self._stringio  # free up some memory
+        sys.stdout = self._stdout
+
+
+# Custom mock for sys.stdin that supports buffer attribute
+class MockStdinWithBuffer:
+    def __init__(self, inputs: str):
+        self.inputs = inputs
+        self._stringio = StringIO(inputs)
+        self.buffer = MockBuffer(inputs)
+
+    def read(self, *args):
+        return self.inputs
+
+    def readline(self, *args):
+        return self._stringio.readline(*args)
+
+    def readlines(self, *args):
+        return self.inputs.split("\n")
+
+    def __getattr__(self, name):
+        # Delegate other attributes to StringIO
+        return getattr(self._stringio, name)
+
+
+class MockBuffer:
+    def __init__(self, inputs: str):
+        self.inputs = inputs.encode("utf-8")  # Convert to bytes
+
+    def read(self, *args):
+        # Return as byte strings that can be split
+        return self.inputs
+
+    def readline(self, *args):
+        return self.inputs.split(b"\n")[0] + b"\n"
+
+
+def clean_if_name(code: str) -> str:
+    try:
+        astree = ast.parse(code)
+        last_block = astree.body[-1]
+        if isinstance(last_block, ast.If):
+            condition = last_block.test
+            if ast.unparse(condition).strip() == "__name__ == '__main__'":
+                code = (
+                    ast.unparse(astree.body[:-1]) + "\n" + ast.unparse(last_block.body)  # type: ignore
+                )
+    except:
+        pass
+
+    return code
+
+
+def make_function(code: str) -> str:
+    try:
+        import_stmts = []
+        all_other_stmts = []
+        astree = ast.parse(code)
+        for stmt in astree.body:
+            if isinstance(stmt, (ast.Import, ast.ImportFrom)):
+                import_stmts.append(stmt)
+            else:
+                all_other_stmts.append(stmt)
+
+        function_ast = ast.FunctionDef(
+            name="wrapped_function",
+            args=ast.arguments(
+                posonlyargs=[], args=[], kwonlyargs=[], kw_defaults=[], defaults=[]
+            ),
+            body=all_other_stmts,
+            decorator_list=[],
+            lineno=-1,
+        )
+        main_code = (
+            import_string
+            + "\n"
+            + ast.unparse(import_stmts)
+            + "\n"
+            + ast.unparse(function_ast)
+        )
+        return main_code
+    except Exception:
+        return code
+
+
+def call_method(method, inputs):
+    if isinstance(inputs, list):
+        inputs = "\n".join(inputs)
+
+    inputs_line_iterator = iter(inputs.split("\n"))
+
+    # Create custom stdin mock with buffer support
+    mock_stdin = MockStdinWithBuffer(inputs)
+
+    # sys.setrecursionlimit(10000)
+
+    # @patch('builtins.input', side_effect=inputs.split("\n"))
+    @patch("builtins.open", mock_open(read_data=inputs))
+    @patch("sys.stdin", mock_stdin)  # Use our custom mock instead of StringIO
+    @patch("sys.stdin.readline", lambda *args: next(inputs_line_iterator))
+    @patch("sys.stdin.readlines", lambda *args: inputs.split("\n"))
+    @patch("sys.stdin.read", lambda *args: inputs)
+    # @patch('sys.stdout.write', print)
+    def _inner_call_method(_method):
+        try:
+            return _method()
+        except SystemExit:
+            pass
+        finally:
+            pass
+
+    return _inner_call_method(method)
+
+
+def get_function(compiled_sol, fn_name: str):  # type: ignore
+    try:
+        assert hasattr(compiled_sol, fn_name)
+        return getattr(compiled_sol, fn_name)
+    except Exception:
+        return
+
+
+def compile_code(code: str, timeout: int):
+    signal.alarm(timeout)
+    try:
+        tmp_sol = ModuleType("tmp_sol", "")
+        exec(code, tmp_sol.__dict__)
+        if "class Solution" in code:
+            # leetcode wraps solutions in `Solution`
+            # this is a hack to check if it is leetcode solution or not
+            # currently livecodebench only supports LeetCode but
+            # else condition allows future extensibility to other platforms
+            compiled_sol = tmp_sol.Solution()
+        else:
+            # do nothing in the other case since function is accesible
+            compiled_sol = tmp_sol
+
+        assert compiled_sol is not None
+    finally:
+        signal.alarm(0)
+
+    return compiled_sol
+
+
+def convert_line_to_decimals(line: str) -> tuple[bool, list[Decimal]]:
+    try:
+        decimal_line = [Decimal(elem) for elem in line.split()]
+    except:
+        return False, []
+    return True, decimal_line
+
+
+def get_stripped_lines(val: str):
+    ## you don't want empty lines to add empty list after splitlines!
+    val = val.strip()
+
+    return [val_line.strip() for val_line in val.split("\n")]
+
+
+def grade_call_based(
+    code: str, all_inputs: list, all_outputs: list, fn_name: str, timeout: int
+):
+    # call-based clean up logic
+    # need to wrap in try-catch logic after to catch the correct errors, but for now this is fine.
+    code = import_string + "\n\n" + code
+    compiled_sol = compile_code(code, timeout)
+
+    if compiled_sol is None:
+        return
+
+    method = get_function(compiled_sol, fn_name)
+
+    if method is None:
+        return
+
+    all_inputs = [
+        [json.loads(line) for line in inputs.split("\n")] for inputs in all_inputs
+    ]
+
+    all_outputs = [json.loads(output) for output in all_outputs]
+
+    total_execution = 0
+    all_results = []
+    for idx, (gt_inp, gt_out) in enumerate(zip(all_inputs, all_outputs)):
+        signal.alarm(timeout)
+        faulthandler.enable()
+        try:
+            # can lock here so time is useful
+            start = time.time()
+            prediction = method(*gt_inp)
+            total_execution += time.time() - start
+            signal.alarm(0)
+
+            # don't penalize model if it produces tuples instead of lists
+            # ground truth sequences are not tuples
+            if isinstance(prediction, tuple):
+                prediction = list(prediction)
+
+            tmp_result = prediction == gt_out
+
+            # handle floating point comparisons
+
+            all_results.append(tmp_result)
+
+            if not tmp_result:
+                return all_results, {
+                    "output": truncatefn(prediction),
+                    "inputs": truncatefn(gt_inp),
+                    "expected": truncatefn(gt_out),
+                    "error_code": -2,
+                    "error_message": "Wrong Answer",
+                }
+        except Exception as e:
+            signal.alarm(0)
+            if "timeoutexception" in repr(e).lower():
+                all_results.append(-3)
+                return all_results, {
+                    "error": repr(e),
+                    "error_code": -3,
+                    "error_message": "Time Limit Exceeded",
+                    "inputs": truncatefn(gt_inp),
+                    "expected": truncatefn(gt_out),
+                }
+            else:
+                all_results.append(-4)
+                return all_results, {
+                    "error": repr(e),
+                    "error_code": -4,
+                    "error_message": "Runtime Error",
+                    "inputs": truncatefn(gt_inp),
+                    "expected": truncatefn(gt_out),
+                }
+
+        finally:
+            signal.alarm(0)
+            faulthandler.disable()
+
+    return all_results, {"execution time": total_execution}
+
+
+def grade_stdio(
+    code: str,
+    all_inputs: list,
+    all_outputs: list,
+    timeout: int,
+):
+    ## runtime doesn't interact well with __name__ == '__main__'
+    code = clean_if_name(code)
+
+    ## we wrap the given code inside another function
+    code = make_function(code)
+
+    compiled_sol = compile_code(code, timeout)
+    if compiled_sol is None:
+        return
+
+    method = get_function(compiled_sol, "wrapped_function")
+
+    if method is None:
+        return
+
+    all_results = []
+    total_execution_time = 0
+    for idx, (gt_inp, gt_out) in enumerate(zip(all_inputs, all_outputs)):
+        signal.alarm(timeout)
+        faulthandler.enable()
+
+        signal.alarm(timeout)
+        with Capturing() as captured_output:
+            try:
+                start = time.time()
+                call_method(method, gt_inp)
+                total_execution_time += time.time() - start
+                # reset the alarm
+                signal.alarm(0)
+            except Exception as e:
+                signal.alarm(0)
+                if "timeoutexception" in repr(e).lower():
+                    all_results.append(-3)
+                    return all_results, {
+                        "error": repr(e),
+                        "error_code": -3,
+                        "error_message": "Time Limit Exceeded",
+                        "inputs": truncatefn(gt_inp),
+                        "expected": truncatefn(gt_out),
+                    }
+                else:
+                    all_results.append(-4)
+                    return all_results, {
+                        "error": repr(e),
+                        "error_code": -4,
+                        "error_message": "Runtime Error",
+                        "inputs": truncatefn(gt_inp),
+                        "expected": truncatefn(gt_out),
+                    }
+
+            finally:
+                signal.alarm(0)
+                faulthandler.disable()
+
+        prediction = captured_output[0]
+
+        stripped_prediction_lines = get_stripped_lines(prediction)
+        stripped_gt_out_lines = get_stripped_lines(gt_out)
+
+        ## WA happens in multiple circumstances
+        ## so cache the return to make it clean!
+        WA_send_args = {
+            "output": truncatefn(prediction),
+            "inputs": truncatefn(gt_inp),
+            "expected": truncatefn(gt_out),
+            "error_code": -2,
+        }
+
+        if len(stripped_prediction_lines) != len(stripped_gt_out_lines):
+            all_results.append(-2)
+            WA_send_args["error_message"] = "Wrong answer: mismatched output length"
+            return all_results, WA_send_args
+
+        for output_line_idx, (
+            stripped_prediction_line,
+            stripped_gt_out_line,
+        ) in enumerate(zip(stripped_prediction_lines, stripped_gt_out_lines)):
+            WA_send_args["error_message"] = (
+                f"Wrong answer at {output_line_idx=}: {truncatefn(stripped_prediction_line)} != {truncatefn(stripped_gt_out_line)}"
+            )
+
+            ## CASE 1: exact match
+            if stripped_prediction_line == stripped_gt_out_line:
+                continue
+
+            ## CASE 2: element-wise comparision
+            ## if there are floating elements
+            ## use `decimal` library for good floating point comparision
+            ## otherwise gotcha: np.isclose(50000000000000000, 50000000000000001) = True
+            ## note that we should always be able to convert to decimals
+
+            success, decimal_prediction_line = convert_line_to_decimals(
+                stripped_prediction_line
+            )
+            if not success:
+                all_results.append(-2)
+                return all_results, WA_send_args
+            success, decimal_gtout_line = convert_line_to_decimals(stripped_gt_out_line)
+            if not success:
+                all_results.append(-2)
+                return all_results, WA_send_args
+
+            if decimal_prediction_line == decimal_gtout_line:
+                continue
+
+            all_results.append(-2)
+            return all_results, WA_send_args
+        all_results.append(True)
+
+    return all_results, {"execution time": total_execution_time}
+
+
+def run_test(sample, test=None, debug=False, timeout=6):
+    """
+    if test(generated_code) is not None it'll try to run the code.
+    otherwise it'll just return an input and output pair.
+    """
+    signal.signal(signal.SIGALRM, timeout_handler)
+
+    # Disable functionalities that can make destructive changes to the test.
+    # max memory is set to 4GB
+    reliability_guard()
+
+    if debug:
+        print(f"start = {datetime.now().time()}")
+
+    try:
+        in_outs = json.loads(sample["input_output"])
+    except ValueError as e:
+        raise e
+        in_outs = None
+
+    if in_outs:
+        if in_outs.get("fn_name") is None:
+            which_type = CODE_TYPE.standard_input  # Standard input
+            method_name = None
+
+        else:
+            which_type = CODE_TYPE.call_based  # Call-based
+            method_name = in_outs["fn_name"]
+
+    if debug:
+        print(f"loaded input_output = {datetime.now().time()}")
+
+    if test is None:
+        assert False, "should not happen: test code is none"
+        return in_outs, {"error": "no test code provided"}
+    elif test is not None:
+        results = []
+        sol = import_string
+        if debug:
+            print(f"loading test code = {datetime.now().time()}")
+
+        if which_type == CODE_TYPE.call_based:
+            signal.alarm(timeout)
+            try:
+                results, metadata = grade_call_based(
+                    code=test,
+                    all_inputs=in_outs["inputs"],
+                    all_outputs=in_outs["outputs"],
+                    fn_name=method_name,
+                    timeout=timeout,
+                )
+                return results, metadata
+            except Exception as e:
+                return [-4], {
+                    "error_code": -4,
+                    "error_message": f"Error during testing: {e}",
+                }
+            finally:
+                signal.alarm(0)
+        elif which_type == CODE_TYPE.standard_input:
+            # sol
+            # if code has if __name__ == "__main__": then remove it
+
+            signal.alarm(timeout)
+            try:
+                results, metadata = grade_stdio(
+                    code=test,
+                    all_inputs=in_outs["inputs"],
+                    all_outputs=in_outs["outputs"],
+                    timeout=timeout,
+                )
+                return results, metadata
+            except Exception as e:
+                return [-4], {
+                    "error_code": -4,
+                    "error_message": f"Error during testing: {e}",
+                }
+            finally:
+                signal.alarm(0)
+
+
+def reliability_guard(maximum_memory_bytes=None):
+    """
+    This disables various destructive functions and prevents the generated code
+    from interfering with the test (e.g. fork bomb, killing other processes,
+    removing filesystem files, etc.)
+    WARNING
+    This function is NOT a security sandbox. Untrusted code, including, model-
+    generated code, should not be blindly executed outside of one. See the
+    Codex paper for more information about OpenAI's code sandbox, and proceed
+    with caution.
+    """
+
+    if maximum_memory_bytes is not None:
+        import resource
+
+        resource.setrlimit(
+            resource.RLIMIT_AS, (maximum_memory_bytes, maximum_memory_bytes)
+        )
+        resource.setrlimit(
+            resource.RLIMIT_DATA, (maximum_memory_bytes, maximum_memory_bytes)
+        )
+        if not platform.uname().system == "Darwin":
+            resource.setrlimit(
+                resource.RLIMIT_STACK, (maximum_memory_bytes, maximum_memory_bytes)
+            )
+
+    faulthandler.disable()
+
+    import builtins
+
+    # builtins.exit = None
+    builtins.quit = None
+
+    import os
+
+    os.environ["OMP_NUM_THREADS"] = "1"
+
+    os.kill = None
+    os.system = None
+    os.putenv = None
+    os.remove = None
+    os.removedirs = None
+    os.rmdir = None
+    os.fchdir = None
+    os.setuid = None
+    os.fork = None
+    os.forkpty = None
+    os.killpg = None
+    os.rename = None
+    os.renames = None
+    os.truncate = None
+    os.replace = None
+    os.unlink = None
+    os.fchmod = None
+    os.fchown = None
+    os.chmod = None
+    os.chown = None
+    os.chroot = None
+    os.fchdir = None
+    os.lchflags = None
+    os.lchmod = None
+    os.lchown = None
+    os.getcwd = None
+    os.chdir = None
+
+    import shutil
+
+    shutil.rmtree = None
+    shutil.move = None
+    shutil.chown = None
+
+    import subprocess
+
+    subprocess.Popen = None
+
+    __builtins__["help"] = None
+
+    import sys
+
+    sys.modules["ipdb"] = None
+    sys.modules["joblib"] = None
+    sys.modules["resource"] = None
+    sys.modules["psutil"] = None
+    sys.modules["tkinter"] = None

dashboard/src/app.css

@@ -202,6 +202,23 @@
 	filter: drop-shadow(0 0 3px oklch(0.85 0.18 85 / 0.5));
 }
 
+/* Onboarding step 2: connection line between devices */
+.onboarding-connection-line {
+	stroke: oklch(0.85 0.18 85 / 0.5);
+	stroke-width: 1.5px;
+	stroke-dasharray: 6, 6;
+	animation: flowAnimation 1s linear infinite;
+	filter: drop-shadow(0 0 4px oklch(0.85 0.18 85 / 0.4));
+}
+
+/* Onboarding step 4: red connection line for disconnect */
+.onboarding-connection-line-red {
+	stroke: rgba(220, 38, 38, 0.7);
+	stroke-width: 1.5px;
+	stroke-dasharray: 6, 6;
+	filter: drop-shadow(0 0 2px rgba(220, 38, 38, 0.3));
+}
+
 .graph-link-active {
 	stroke: oklch(0.85 0.18 85 / 0.8);
 	stroke-width: 2px;
@@ -320,3 +337,51 @@ input:focus, textarea:focus {
 		transform: translate(400px, 400px);
 	}
 }
+
+/* Onboarding smooth fade-in */
+@keyframes onb-fade-in {
+	from { opacity: 0; transform: translateY(8px); }
+	to { opacity: 1; transform: none; }
+}
+
+/* Opacity-only fade-in — no transform, safe for containers with fixed-position children */
+@keyframes onb-fade-opacity {
+	from { opacity: 0; }
+	to { opacity: 1; }
+}
+
+/* Respect reduced motion preference */
+@media (prefers-reduced-motion: reduce) {
+	.shooting-star,
+	.shooting-star::before {
+		animation: none !important;
+		opacity: 0 !important;
+	}
+	.graph-link {
+		animation: none;
+	}
+	.status-pulse {
+		animation: none;
+	}
+	.cursor-blink {
+		animation: none;
+	}
+	.onboarding-connection-line {
+		animation: none;
+	}
+	.onboarding-connection-line-red {
+		animation: none;
+	}
+	[style*="onb-fade-in"],
+	[style*="onb-fade-opacity"] {
+		animation: none !important;
+		opacity: 1 !important;
+	}
+	*,
+	*::before,
+	*::after {
+		transition-duration: 0.01ms !important;
+		animation-duration: 0.01ms !important;
+		animation-iteration-count: 1 !important;
+	}
+}

dashboard/src/lib/components/ChatForm.svelte

@@ -1,19 +1,15 @@
 <script lang="ts">
   import {
     isLoading,
-    sendMessage,
-    generateImage,
-    editImage,
     editingImage,
     clearEditingImage,
     selectedChatModel,
-    setSelectedChatModel,
-    instances,
     ttftMs,
     tps,
     totalTokens,
     thinkingEnabled as thinkingEnabledStore,
     setConversationThinking,
+    stopGeneration,
   } from "$lib/stores/app.svelte";
   import ChatAttachments from "./ChatAttachments.svelte";
   import ImageParamsPanel from "./ImageParamsPanel.svelte";
@@ -28,6 +24,19 @@
     showModelSelector?: boolean;
     modelTasks?: Record<string, string[]>;
     modelCapabilities?: Record<string, string[]>;
+    onSend?: () => void;
+    onAutoSend: (
+      content: string,
+      files?: {
+        id: string;
+        name: string;
+        type: string;
+        textContent?: string;
+        preview?: string;
+      }[],
+    ) => void;
+    onOpenModelPicker?: () => void;
+    modelDisplayOverride?: string;
   }
 
   let {
@@ -38,6 +47,10 @@
     showModelSelector = false,
     modelTasks = {},
     modelCapabilities = {},
+    onSend,
+    onAutoSend,
+    onOpenModelPicker,
+    modelDisplayOverride,
   }: Props = $props();
 
   let message = $state("");
@@ -48,27 +61,12 @@
   const thinkingEnabled = $derived(thinkingEnabledStore());
   let loading = $derived(isLoading());
   const currentModel = $derived(selectedChatModel());
-  const instanceData = $derived(instances());
   const currentTtft = $derived(ttftMs());
   const currentTps = $derived(tps());
   const currentTokens = $derived(totalTokens());
   const currentEditingImage = $derived(editingImage());
   const isEditMode = $derived(currentEditingImage !== null);
 
-  // Custom dropdown state
-  let isModelDropdownOpen = $state(false);
-  let dropdownButtonRef: HTMLButtonElement | undefined = $state();
-  let dropdownPosition = $derived(() => {
-    if (!dropdownButtonRef || !isModelDropdownOpen)
-      return { top: 0, left: 0, width: 0 };
-    const rect = dropdownButtonRef.getBoundingClientRect();
-    return {
-      top: rect.top,
-      left: rect.left,
-      width: rect.width,
-    };
-  });
-
   // Accept all supported file types
   const acceptString = getAcceptString(["image", "text", "pdf"]);
 
@@ -103,7 +101,7 @@
   const modelSupportsThinking = $derived(() => {
     if (!currentModel) return false;
     const caps = modelCapabilities[currentModel] || [];
-    return caps.includes("thinking") && caps.includes("text");
+    return caps.includes("thinking_toggle") && caps.includes("text");
   });
 
   const isEditOnlyWithoutImage = $derived(
@@ -121,73 +119,14 @@
         uploadedFiles.length > 0),
   );
 
-  // Extract available models from running instances
-  const availableModels = $derived(() => {
-    const models: Array<{ id: string; label: string; isImageModel: boolean }> =
-      [];
-    for (const [, instance] of Object.entries(instanceData)) {
-      const modelId = getInstanceModelId(instance);
-      if (
-        modelId &&
-        modelId !== "Unknown" &&
-        !models.some((m) => m.id === modelId)
-      ) {
-        models.push({
-          id: modelId,
-          label: modelId.split("/").pop() || modelId,
-          isImageModel: modelSupportsImageGeneration(modelId),
-        });
-      }
-    }
-    return models;
-  });
-
-  // Track previous model IDs to detect newly added models (plain variable to avoid reactive loop)
-  let previousModelIds: Set<string> = new Set();
-
-  // Auto-select the first available model if none is selected, if current selection is stale, or if a new model is added
-  $effect(() => {
-    const models = availableModels();
-    const currentModelIds = new Set(models.map((m) => m.id));
-
-    if (models.length > 0) {
-      // Find newly added models (in current but not in previous)
-      const newModels = models.filter((m) => !previousModelIds.has(m.id));
-
-      // If no model selected, select the first available
-      if (!currentModel) {
-        setSelectedChatModel(models[0].id);
-      }
-      // If current model is stale (no longer has a running instance), reset to first available
-      else if (!models.some((m) => m.id === currentModel)) {
-        setSelectedChatModel(models[0].id);
-      }
-      // If a new model was just added, select it
-      else if (newModels.length > 0 && previousModelIds.size > 0) {
-        setSelectedChatModel(newModels[0].id);
-      }
-    } else {
-      // No instances running - clear the selected model
-      if (currentModel) {
-        setSelectedChatModel("");
-      }
-    }
-
-    // Update previous model IDs for next comparison
-    previousModelIds = currentModelIds;
-  });
-
-  function getInstanceModelId(instanceWrapped: unknown): string {
-    if (!instanceWrapped || typeof instanceWrapped !== "object") return "";
-    const keys = Object.keys(instanceWrapped as Record<string, unknown>);
-    if (keys.length === 1) {
-      const instance = (instanceWrapped as Record<string, unknown>)[
-        keys[0]
-      ] as { shardAssignments?: { modelId?: string } };
-      return instance?.shardAssignments?.modelId || "";
-    }
-    return "";
-  }
+  // Short label for the currently selected model
+  const currentModelLabel = $derived(
+    currentModel
+      ? currentModel.split("/").pop() || currentModel
+      : modelDisplayOverride
+        ? modelDisplayOverride.split("/").pop() || modelDisplayOverride
+        : "",
+  );
 
   async function handleFiles(files: File[]) {
     if (files.length === 0) return;
@@ -274,38 +213,10 @@
     uploadedFiles = [];
     resetTextareaHeight();
 
-    // Use image editing if in edit mode
-    if (isEditMode && currentEditingImage && content) {
-      editImage(content, currentEditingImage.imageDataUrl);
-    }
-    // If user attached an image with an ImageToImage model, use edit endpoint
-    else if (
-      currentModel &&
-      modelSupportsImageEditing(currentModel) &&
-      files.length > 0 &&
-      content
-    ) {
-      // Use the first attached image for editing
-      const imageFile = files[0];
-      if (imageFile.preview) {
-        editImage(content, imageFile.preview);
-      }
-    } else if (
-      currentModel &&
-      modelSupportsTextToImage(currentModel) &&
-      content
-    ) {
-      // Use image generation for text-to-image models
-      generateImage(content);
-    } else {
-      sendMessage(
-        content,
-        files,
-        modelSupportsThinking() ? thinkingEnabled : null,
-      );
-    }
-
-    // Refocus the textarea after sending
+    // Parent controls all send logic (including image routing,
+    // launching non-running models before sending, etc.)
+    onAutoSend(content, files);
+    onSend?.();
     setTimeout(() => textareaRef?.focus(), 10);
   }
 
@@ -415,7 +326,7 @@
     {/if}
 
     <!-- Model selector (when enabled) -->
-    {#if showModelSelector && availableModels().length > 0}
+    {#if showModelSelector}
       <div
         class="flex items-center justify-between gap-2 px-3 py-2 border-b border-exo-medium-gray/30"
       >
@@ -424,33 +335,22 @@
             class="text-xs text-exo-light-gray uppercase tracking-wider flex-shrink-0"
             >MODEL:</span
           >
-          <!-- Custom dropdown -->
+          <!-- Model button — opens the full model picker -->
           <div class="relative flex-1 max-w-xs">
             <button
-              bind:this={dropdownButtonRef}
               type="button"
-              onclick={() => (isModelDropdownOpen = !isModelDropdownOpen)}
-              class="w-full bg-exo-medium-gray/50 border border-exo-yellow/30 rounded pl-3 pr-8 py-1.5 text-xs font-mono text-left tracking-wide cursor-pointer transition-all duration-200 hover:border-exo-yellow/50 focus:outline-none focus:border-exo-yellow/70 {isModelDropdownOpen
-                ? 'border-exo-yellow/70'
-                : ''}"
+              onclick={() => onOpenModelPicker?.()}
+              class="w-full bg-exo-medium-gray/50 border border-exo-yellow/30 rounded pl-3 pr-8 py-1.5 text-xs font-mono text-left tracking-wide cursor-pointer transition-all duration-200 hover:border-exo-yellow/50 focus:outline-none focus:border-exo-yellow/70"
             >
-              {#if availableModels().find((m) => m.id === currentModel)}
-                <span class="text-exo-yellow truncate"
-                  >{availableModels().find((m) => m.id === currentModel)
-                    ?.label}</span
-                >
-              {:else if availableModels().length > 0}
-                <span class="text-exo-yellow truncate"
-                  >{availableModels()[0].label}</span
+              {#if currentModelLabel}
+                <span class="text-exo-yellow truncate">{currentModelLabel}</span
                 >
               {:else}
                 <span class="text-exo-light-gray/50">— SELECT MODEL —</span>
               {/if}
             </button>
             <div
-              class="absolute right-2 top-1/2 -translate-y-1/2 pointer-events-none transition-transform duration-200 {isModelDropdownOpen
-                ? 'rotate-180'
-                : ''}"
+              class="absolute right-2 top-1/2 -translate-y-1/2 pointer-events-none"
             >
               <svg
                 class="w-3 h-3 text-exo-yellow/60"
@@ -467,78 +367,6 @@
               </svg>
             </div>
           </div>
-
-          {#if isModelDropdownOpen}
-            <!-- Backdrop to close dropdown -->
-            <button
-              type="button"
-              class="fixed inset-0 z-[9998] cursor-default"
-              onclick={() => (isModelDropdownOpen = false)}
-              aria-label="Close dropdown"
-            ></button>
-
-            <!-- Dropdown Panel - fixed positioning to escape overflow:hidden -->
-            <div
-              class="fixed bg-exo-dark-gray border border-exo-yellow/30 rounded shadow-lg shadow-black/50 z-[9999] max-h-48 overflow-y-auto"
-              style="bottom: calc(100vh - {dropdownPosition()
-                .top}px + 4px); left: {dropdownPosition()
-                .left}px; width: {dropdownPosition().width}px;"
-            >
-              <div class="py-1">
-                {#each availableModels() as model}
-                  <button
-                    type="button"
-                    onclick={() => {
-                      setSelectedChatModel(model.id);
-                      isModelDropdownOpen = false;
-                    }}
-                    class="w-full px-3 py-2 text-left text-xs font-mono tracking-wide transition-colors duration-100 flex items-center gap-2 {currentModel ===
-                    model.id
-                      ? 'bg-transparent text-exo-yellow'
-                      : 'text-exo-light-gray hover:text-exo-yellow'}"
-                  >
-                    {#if currentModel === model.id}
-                      <svg
-                        class="w-3 h-3 flex-shrink-0"
-                        fill="currentColor"
-                        viewBox="0 0 20 20"
-                      >
-                        <path
-                          fill-rule="evenodd"
-                          d="M16.707 5.293a1 1 0 010 1.414l-8 8a1 1 0 01-1.414 0l-4-4a1 1 0 011.414-1.414L8 12.586l7.293-7.293a1 1 0 011.414 0z"
-                          clip-rule="evenodd"
-                        />
-                      </svg>
-                    {:else}
-                      <span class="w-3"></span>
-                    {/if}
-                    {#if model.isImageModel}
-                      <svg
-                        class="w-3.5 h-3.5 flex-shrink-0 text-exo-yellow"
-                        fill="none"
-                        viewBox="0 0 24 24"
-                        stroke="currentColor"
-                        stroke-width="2"
-                        aria-label="Image generation model"
-                      >
-                        <rect
-                          x="3"
-                          y="3"
-                          width="18"
-                          height="18"
-                          rx="2"
-                          ry="2"
-                        />
-                        <circle cx="8.5" cy="8.5" r="1.5" />
-                        <polyline points="21 15 16 10 5 21" />
-                      </svg>
-                    {/if}
-                    <span class="truncate flex-1">{model.label}</span>
-                  </button>
-                {/each}
-              </div>
-            </div>
-          {/if}
         </div>
         <!-- Thinking toggle -->
         {#if modelSupportsThinking()}
@@ -653,86 +481,92 @@
         style="min-height: 28px; max-height: 150px;"
       ></textarea>
 
-      <button
-        type="submit"
-        disabled={!canSend || loading || isEditOnlyWithoutImage}
-        class="px-2.5 sm:px-4 py-1.5 sm:py-2 rounded text-xs sm:text-xs tracking-[0.1em] sm:tracking-[0.15em] uppercase font-medium transition-all duration-200 whitespace-nowrap
-					{!canSend || loading || isEditOnlyWithoutImage
-          ? 'bg-exo-medium-gray/50 text-exo-light-gray cursor-not-allowed'
-          : 'bg-exo-yellow text-exo-black hover:bg-exo-yellow-darker hover:shadow-[0_0_20px_rgba(255,215,0,0.3)]'}"
-        aria-label={shouldShowEditMode
-          ? "Edit image"
-          : isImageModel()
-            ? "Generate image"
-            : "Send message"}
-      >
-        {#if loading}
+      {#if loading}
+        <button
+          type="button"
+          onclick={() => stopGeneration()}
+          class="px-2.5 sm:px-4 py-1.5 sm:py-2 rounded text-xs sm:text-xs tracking-[0.1em] sm:tracking-[0.15em] font-medium transition-all duration-200 whitespace-nowrap bg-exo-medium-gray/70 text-exo-light-gray hover:bg-exo-medium-gray hover:text-white"
+          aria-label="Stop generation"
+        >
           <span class="inline-flex items-center gap-1 sm:gap-2">
-            <span
-              class="w-2.5 h-2.5 sm:w-3 sm:h-3 border-2 border-current border-t-transparent rounded-full animate-spin"
-            ></span>
-            <span class="hidden sm:inline"
-              >{shouldShowEditMode
-                ? "EDITING"
-                : isImageModel()
-                  ? "GENERATING"
-                  : "PROCESSING"}</span
-            >
-            <span class="sm:hidden">...</span>
-          </span>
-        {:else if shouldShowEditMode}
-          <span class="inline-flex items-center gap-1.5">
             <svg
-              class="w-3.5 h-3.5"
-              fill="none"
+              class="w-3 h-3 sm:w-3.5 sm:h-3.5"
+              fill="currentColor"
               viewBox="0 0 24 24"
-              stroke="currentColor"
-              stroke-width="2"
             >
-              <path
-                stroke-linecap="round"
-                stroke-linejoin="round"
-                d="M11 5H6a2 2 0 00-2 2v11a2 2 0 002 2h11a2 2 0 002-2v-5m-1.414-9.414a2 2 0 112.828 2.828L11.828 15H9v-2.828l8.586-8.586z"
-              />
+              <rect x="6" y="6" width="12" height="12" rx="1" />
             </svg>
-            <span>EDIT</span>
+            <span class="hidden sm:inline">Cancel</span>
           </span>
-        {:else if isEditOnlyWithoutImage}
-          <span class="inline-flex items-center gap-1.5">
-            <svg
-              class="w-3.5 h-3.5"
-              fill="none"
-              viewBox="0 0 24 24"
-              stroke="currentColor"
-              stroke-width="2"
-            >
-              <path
-                stroke-linecap="round"
-                stroke-linejoin="round"
-                d="M11 5H6a2 2 0 00-2 2v11a2 2 0 002 2h11a2 2 0 002-2v-5m-1.414-9.414a2 2 0 112.828 2.828L11.828 15H9v-2.828l8.586-8.586z"
-              />
-            </svg>
-            <span>EDIT</span>
-          </span>
-        {:else if isImageModel()}
-          <span class="inline-flex items-center gap-1.5">
-            <svg
-              class="w-3.5 h-3.5"
-              fill="none"
-              viewBox="0 0 24 24"
-              stroke="currentColor"
-              stroke-width="2"
-            >
-              <rect x="3" y="3" width="18" height="18" rx="2" ry="2" />
-              <circle cx="8.5" cy="8.5" r="1.5" />
-              <polyline points="21 15 16 10 5 21" />
-            </svg>
-            <span>GENERATE</span>
-          </span>
-        {:else}
-          SEND
-        {/if}
-      </button>
+        </button>
+      {:else}
+        <button
+          type="submit"
+          disabled={!canSend || isEditOnlyWithoutImage}
+          class="px-2.5 sm:px-4 py-1.5 sm:py-2 rounded text-xs sm:text-xs tracking-[0.1em] sm:tracking-[0.15em] uppercase font-medium transition-all duration-200 whitespace-nowrap
+					{!canSend || isEditOnlyWithoutImage
+            ? 'bg-exo-medium-gray/50 text-exo-light-gray cursor-not-allowed'
+            : 'bg-exo-yellow text-exo-black hover:bg-exo-yellow-darker hover:shadow-[0_0_20px_rgba(255,215,0,0.3)]'}"
+          aria-label={shouldShowEditMode
+            ? "Edit image"
+            : isImageModel()
+              ? "Generate image"
+              : "Send message"}
+        >
+          {#if shouldShowEditMode}
+            <span class="inline-flex items-center gap-1.5">
+              <svg
+                class="w-3.5 h-3.5"
+                fill="none"
+                viewBox="0 0 24 24"
+                stroke="currentColor"
+                stroke-width="2"
+              >
+                <path
+                  stroke-linecap="round"
+                  stroke-linejoin="round"
+                  d="M11 5H6a2 2 0 00-2 2v11a2 2 0 002 2h11a2 2 0 002-2v-5m-1.414-9.414a2 2 0 112.828 2.828L11.828 15H9v-2.828l8.586-8.586z"
+                />
+              </svg>
+              <span>EDIT</span>
+            </span>
+          {:else if isEditOnlyWithoutImage}
+            <span class="inline-flex items-center gap-1.5">
+              <svg
+                class="w-3.5 h-3.5"
+                fill="none"
+                viewBox="0 0 24 24"
+                stroke="currentColor"
+                stroke-width="2"
+              >
+                <path
+                  stroke-linecap="round"
+                  stroke-linejoin="round"
+                  d="M11 5H6a2 2 0 00-2 2v11a2 2 0 002 2h11a2 2 0 002-2v-5m-1.414-9.414a2 2 0 112.828 2.828L11.828 15H9v-2.828l8.586-8.586z"
+                />
+              </svg>
+              <span>EDIT</span>
+            </span>
+          {:else if isImageModel()}
+            <span class="inline-flex items-center gap-1.5">
+              <svg
+                class="w-3.5 h-3.5"
+                fill="none"
+                viewBox="0 0 24 24"
+                stroke="currentColor"
+                stroke-width="2"
+              >
+                <rect x="3" y="3" width="18" height="18" rx="2" ry="2" />
+                <circle cx="8.5" cy="8.5" r="1.5" />
+                <polyline points="21 15 16 10 5 21" />
+              </svg>
+              <span>GENERATE</span>
+            </span>
+          {:else}
+            SEND
+          {/if}
+        </button>
+      {/if}
     </div>
 
     <!-- Bottom accent line -->

dashboard/src/lib/components/ChatMessages.svelte

@@ -3,16 +3,17 @@
     messages,
     currentResponse,
     isLoading,
+    prefillProgress,
     deleteMessage,
     editAndRegenerate,
     regenerateLastResponse,
     regenerateFromToken,
     setEditingImage,
   } from "$lib/stores/app.svelte";
-  import type { Message } from "$lib/stores/app.svelte";
   import type { MessageAttachment } from "$lib/stores/app.svelte";
   import MarkdownContent from "./MarkdownContent.svelte";
   import TokenHeatmap from "./TokenHeatmap.svelte";
+  import PrefillProgressBar from "./PrefillProgressBar.svelte";
   import ImageLightbox from "./ImageLightbox.svelte";
 
   interface Props {
@@ -25,6 +26,7 @@
   const messageList = $derived(messages());
   const response = $derived(currentResponse());
   const loading = $derived(isLoading());
+  const prefill = $derived(prefillProgress());
 
   // Scroll management - user controls scroll, show button when not at bottom
   const SCROLL_THRESHOLD = 100;
@@ -428,6 +430,9 @@
             {:else}
               <!-- Assistant message styling -->
               <div class="p-3 sm:p-4">
+                {#if loading && isLastAssistantMessage(message.id) && prefill && !message.content}
+                  <PrefillProgressBar progress={prefill} class="mb-3" />
+                {/if}
                 {#if message.thinking && message.thinking.trim().length > 0}
                   <div
                     class="mb-3 rounded border border-exo-yellow/20 bg-exo-black/40"
@@ -802,8 +807,8 @@
       >
         AWAITING INPUT
       </p>
-      <p class="text-sm sm:text-xs text-exo-light-gray tracking-wider mt-1">
-        ENTER A QUERY TO BEGIN
+      <p class="text-xs text-white/30 tracking-wider mt-1.5 font-mono">
+        Type a message below &middot; Shift+Enter for newline
       </p>
     </div>
   {/if}
@@ -818,6 +823,7 @@
       onclick={scrollToBottom}
       class="sticky bottom-4 left-1/2 -translate-x-1/2 w-10 h-10 rounded-full bg-exo-dark-gray/90 border border-exo-medium-gray/50 flex items-center justify-center text-exo-light-gray hover:text-exo-yellow hover:border-exo-yellow/50 transition-all shadow-lg cursor-pointer z-10"
       title="Scroll to bottom"
+      aria-label="Scroll to bottom of messages"
     >
       <svg
         class="w-5 h-5"

dashboard/src/lib/components/ChatModelSelector.svelte

@@ -0,0 +1,401 @@
+<script lang="ts" module>
+  export interface ChatModelInfo {
+    id: string;
+    name: string;
+    base_model: string;
+    storage_size_megabytes: number;
+    capabilities: string[];
+    family: string;
+    quantization: string;
+  }
+
+  // Auto mode tier list (for when user just starts typing)
+  export const AUTO_TIERS: string[][] = [
+    // Tier 1 (frontier)
+    ["DeepSeek V3.1", "GLM-5", "Kimi K2.5", "Qwen3 Coder Next"],
+    // Tier 2 (excellent)
+    [
+      "Kimi K2",
+      "Qwen3 235B",
+      "MiniMax M2.5",
+      "Step 3.5 Flash",
+      "Qwen3 Next 80B",
+    ],
+    // Tier 3 (great)
+    [
+      "GLM 4.7",
+      "MiniMax M2.1",
+      "Qwen3 Coder 480B",
+      "GLM 4.5 Air",
+      "Llama 3.3 70B",
+    ],
+    // Tier 4 (good)
+    ["GPT-OSS 120B", "Qwen3 30B", "Llama 3.1 70B", "GLM 4.7 Flash"],
+    // Tier 5 (small/fast)
+    [
+      "Llama 3.1 8B",
+      "GPT-OSS 20B",
+      "Llama 3.2 3B",
+      "Qwen3 0.6B",
+      "Llama 3.2 1B",
+    ],
+  ];
+
+  /** Return the tier index (0 = best) for a base_model name. */
+  export function getAutoTierIndex(baseModel: string): number {
+    for (let i = 0; i < AUTO_TIERS.length; i++) {
+      if (AUTO_TIERS[i].includes(baseModel)) return i;
+    }
+    return AUTO_TIERS.length; // not in any tier → lowest priority
+  }
+
+  /** Auto mode: walk tiers top-down, pick biggest fitting variant from highest tier. */
+  export function pickAutoModel(
+    modelList: ChatModelInfo[],
+    memoryGB: number,
+  ): ChatModelInfo | null {
+    for (const tier of AUTO_TIERS) {
+      const candidates: ChatModelInfo[] = [];
+      for (const baseModel of tier) {
+        const variants = modelList
+          .filter(
+            (m) =>
+              m.base_model === baseModel &&
+              (m.storage_size_megabytes || 0) / 1024 <= memoryGB &&
+              (m.storage_size_megabytes || 0) > 0,
+          )
+          .sort(
+            (a, b) =>
+              (b.storage_size_megabytes || 0) - (a.storage_size_megabytes || 0),
+          );
+        if (variants[0]) candidates.push(variants[0]);
+      }
+      if (candidates.length > 0) {
+        candidates.sort(
+          (a, b) =>
+            (b.storage_size_megabytes || 0) - (a.storage_size_megabytes || 0),
+        );
+        return candidates[0];
+      }
+    }
+    return null;
+  }
+</script>
+
+<script lang="ts">
+  interface CategoryRecommendation {
+    category: "coding" | "writing" | "agentic" | "biggest";
+    label: string;
+    model: ChatModelInfo | null;
+    tooltip: string;
+  }
+
+  interface Props {
+    models: ChatModelInfo[];
+    clusterLabel: string;
+    totalMemoryGB: number;
+    onSelect: (modelId: string, category: string) => void;
+    onAddModel: () => void;
+    class?: string;
+  }
+
+  let {
+    models,
+    clusterLabel,
+    totalMemoryGB,
+    onSelect,
+    onAddModel,
+    class: className = "",
+  }: Props = $props();
+
+  // --- Hardcoded Rankings ---
+  const CODING_RANKING = [
+    "Qwen3 Coder Next",
+    "Qwen3 Coder 480B",
+    "Qwen3 30B",
+    "GPT-OSS 20B",
+    "Llama 3.1 8B",
+    "Llama 3.2 3B",
+    "Qwen3 0.6B",
+  ];
+
+  const WRITING_RANKING = [
+    "Kimi K2.5",
+    "Kimi K2",
+    "Qwen3 Next 80B",
+    "Llama 3.3 70B",
+    "MiniMax M2.5",
+    "GLM 4.5 Air",
+    "GLM 4.7 Flash",
+    "GPT-OSS 20B",
+    "Llama 3.1 8B",
+    "Llama 3.2 3B",
+    "Qwen3 0.6B",
+  ];
+
+  const AGENTIC_RANKING = [
+    "DeepSeek V3.1",
+    "GLM-5",
+    "Qwen3 235B",
+    "Step 3.5 Flash",
+    "GLM 4.7",
+    "MiniMax M2.1",
+    "GPT-OSS 120B",
+    "Llama 3.3 70B",
+    "Llama 3.1 70B",
+    "GLM 4.7 Flash",
+    "GPT-OSS 20B",
+    "Qwen3 30B",
+    "Llama 3.1 8B",
+    "Llama 3.2 3B",
+    "Qwen3 0.6B",
+  ];
+
+  function getModelSizeGB(m: ChatModelInfo): number {
+    return (m.storage_size_megabytes || 0) / 1024;
+  }
+
+  function fitsInMemory(m: ChatModelInfo): boolean {
+    return getModelSizeGB(m) <= totalMemoryGB && getModelSizeGB(m) > 0;
+  }
+
+  /** For a given base_model name, find the biggest quant variant that fits in memory. */
+  function pickBestVariant(baseModel: string): ChatModelInfo | null {
+    const variants = models
+      .filter((m) => m.base_model === baseModel && fitsInMemory(m))
+      .sort((a, b) => getModelSizeGB(b) - getModelSizeGB(a));
+    return variants[0] ?? null;
+  }
+
+  /** Walk a ranked list of base_model names, return the first that has a fitting variant. */
+  function pickFromRanking(ranking: string[]): ChatModelInfo | null {
+    for (const baseModel of ranking) {
+      const pick = pickBestVariant(baseModel);
+      if (pick) return pick;
+    }
+    return null;
+  }
+
+  /** Pick the single biggest model that fits. */
+  function pickBiggest(): ChatModelInfo | null {
+    const fitting = models
+      .filter((m) => fitsInMemory(m))
+      .sort((a, b) => getModelSizeGB(b) - getModelSizeGB(a));
+    return fitting[0] ?? null;
+  }
+
+  const recommendations = $derived.by((): CategoryRecommendation[] => {
+    return [
+      {
+        category: "coding",
+        label: "Best for Coding",
+        model: pickFromRanking(CODING_RANKING),
+        tooltip:
+          "Ranked by coding benchmark performance (LiveCodeBench, SWE-bench)",
+      },
+      {
+        category: "writing",
+        label: "Best for Writing",
+        model: pickFromRanking(WRITING_RANKING),
+        tooltip: "Ranked by creative writing quality and instruction following",
+      },
+      {
+        category: "agentic",
+        label: "Best Agentic",
+        model: pickFromRanking(AGENTIC_RANKING),
+        tooltip: "Ranked by reasoning, planning, and tool-use capability",
+      },
+      {
+        category: "biggest",
+        label: "Biggest",
+        model: pickBiggest(),
+        tooltip: "Largest model that fits in your available memory",
+      },
+    ];
+  });
+
+  function formatSize(mb: number): string {
+    const gb = mb / 1024;
+    if (gb >= 100) return `${Math.round(gb)} GB`;
+    return `${gb.toFixed(1)} GB`;
+  }
+
+  // Category icons (SVG paths)
+  const categoryIcons: Record<string, string> = {
+    coding:
+      "M8 9l3 3-3 3m5 0h3M5 20h14a2 2 0 002-2V6a2 2 0 00-2-2H5a2 2 0 00-2 2v12a2 2 0 002 2z",
+    writing:
+      "M11 5H6a2 2 0 00-2 2v11a2 2 0 002 2h11a2 2 0 002-2v-5m-1.414-9.414a2 2 0 112.828 2.828L11.828 15H9v-2.828l8.586-8.586z",
+    agentic:
+      "M9.663 17h4.673M12 3v1m6.364 1.636l-.707.707M21 12h-1M4 12H3m3.343-5.657l-.707-.707m2.828 9.9a5 5 0 117.072 0l-.548.547A3.374 3.374 0 0014 18.469V19a2 2 0 11-4 0v-.531c0-.895-.356-1.754-.988-2.386l-.548-.547z",
+    biggest:
+      "M19 11H5m14 0a2 2 0 012 2v6a2 2 0 01-2 2H5a2 2 0 01-2-2v-6a2 2 0 012-2m14 0V9a2 2 0 00-2-2M5 11V9a2 2 0 012-2m0 0V5a2 2 0 012-2h6a2 2 0 012 2v2M7 7h10",
+  };
+
+  let hoveredTooltip = $state<string | null>(null);
+  let tooltipAnchor = $state<{ x: number; y: number } | null>(null);
+
+  function showTooltip(category: string, e: MouseEvent | FocusEvent) {
+    hoveredTooltip = category;
+    const target = e.currentTarget as HTMLElement;
+    const rect = target.getBoundingClientRect();
+    tooltipAnchor = { x: rect.left + rect.width / 2, y: rect.top };
+  }
+
+  function hideTooltip() {
+    hoveredTooltip = null;
+    tooltipAnchor = null;
+  }
+</script>
+
+<div class="flex flex-col items-center justify-center gap-6 {className}">
+  <!-- Header -->
+  <div class="text-center">
+    <p class="text-xs text-exo-light-gray uppercase tracking-[0.2em] mb-1">
+      Recommended for your
+    </p>
+    <p class="text-sm text-white font-mono tracking-wide">{clusterLabel}</p>
+  </div>
+
+  <!-- Category Cards Grid -->
+  <div class="grid grid-cols-2 gap-3 w-full max-w-md">
+    {#each recommendations as rec}
+      {#if rec.model}
+        <button
+          type="button"
+          onclick={() => rec.model && onSelect(rec.model.id, rec.category)}
+          class="group relative flex flex-col items-start gap-2 p-4 rounded-lg border border-exo-medium-gray/50 bg-exo-dark-gray/50 hover:border-exo-yellow/40 hover:bg-exo-dark-gray transition-all duration-200 cursor-pointer text-left"
+        >
+          <!-- Category icon + label -->
+          <div class="flex items-center gap-2 w-full">
+            <svg
+              class="w-4 h-4 text-exo-yellow/70 group-hover:text-exo-yellow transition-colors flex-shrink-0"
+              fill="none"
+              viewBox="0 0 24 24"
+              stroke="currentColor"
+              stroke-width="1.5"
+            >
+              <path
+                stroke-linecap="round"
+                stroke-linejoin="round"
+                d={categoryIcons[rec.category]}
+              />
+            </svg>
+            <span
+              class="text-xs font-mono uppercase tracking-wider text-exo-light-gray group-hover:text-white transition-colors"
+            >
+              {rec.label}
+            </span>
+            <!-- Info tooltip -->
+            <div class="ml-auto flex-shrink-0">
+              <span
+                role="button"
+                tabindex="-1"
+                class="text-exo-light-gray/40 hover:text-exo-light-gray transition-colors cursor-help inline-flex"
+                onmouseenter={(e: MouseEvent) => showTooltip(rec.category, e)}
+                onmouseleave={() => hideTooltip()}
+                onclick={(e: MouseEvent) => {
+                  e.stopPropagation();
+                  if (hoveredTooltip === rec.category) {
+                    hideTooltip();
+                  } else {
+                    showTooltip(rec.category, e);
+                  }
+                }}
+              >
+                <svg
+                  class="w-3.5 h-3.5"
+                  fill="none"
+                  viewBox="0 0 24 24"
+                  stroke="currentColor"
+                  stroke-width="2"
+                >
+                  <circle cx="12" cy="12" r="10" />
+                  <path d="M12 16v-4m0-4h.01" />
+                </svg>
+              </span>
+            </div>
+          </div>
+
+          <!-- Model name + size -->
+          <div class="w-full">
+            <p class="text-sm text-white font-mono truncate">
+              {rec.model.base_model}
+            </p>
+            <p class="text-xs text-exo-light-gray/60 font-mono mt-0.5">
+              {formatSize(rec.model.storage_size_megabytes)}
+              {#if rec.model.quantization}
+                <span class="text-exo-light-gray/40"
+                  >&middot; {rec.model.quantization}</span
+                >
+              {/if}
+            </p>
+          </div>
+        </button>
+      {:else}
+        <!-- No model fits for this category -->
+        <div
+          class="flex flex-col items-start gap-2 p-4 rounded-lg border border-exo-medium-gray/30 bg-exo-dark-gray/30 opacity-50"
+        >
+          <div class="flex items-center gap-2">
+            <svg
+              class="w-4 h-4 text-exo-light-gray/40 flex-shrink-0"
+              fill="none"
+              viewBox="0 0 24 24"
+              stroke="currentColor"
+              stroke-width="1.5"
+            >
+              <path
+                stroke-linecap="round"
+                stroke-linejoin="round"
+                d={categoryIcons[rec.category]}
+              />
+            </svg>
+            <span
+              class="text-xs font-mono uppercase tracking-wider text-exo-light-gray/50"
+              >{rec.label}</span
+            >
+          </div>
+          <p class="text-xs text-exo-light-gray/40 font-mono">No model fits</p>
+        </div>
+      {/if}
+    {/each}
+  </div>
+
+  <!-- Add Model Button -->
+  <button
+    type="button"
+    onclick={onAddModel}
+    class="flex items-center gap-2 px-4 py-2 text-xs font-mono uppercase tracking-wider text-exo-light-gray hover:text-exo-yellow border border-exo-medium-gray/30 hover:border-exo-yellow/30 rounded-lg transition-all duration-200 cursor-pointer"
+  >
+    <svg
+      class="w-3.5 h-3.5"
+      fill="none"
+      viewBox="0 0 24 24"
+      stroke="currentColor"
+      stroke-width="2"
+    >
+      <path stroke-linecap="round" stroke-linejoin="round" d="M12 4v16m8-8H4" />
+    </svg>
+    Add Model
+  </button>
+
+  <!-- Auto hint -->
+  <p class="text-xs text-exo-light-gray/40 font-mono tracking-wide text-center">
+    Or just start typing &mdash; we'll pick the best model automatically
+  </p>
+</div>
+
+<!-- Fixed-position tooltip (escapes overflow-hidden ancestors) -->
+{#if hoveredTooltip && tooltipAnchor}
+  {@const rec = recommendations.find((r) => r.category === hoveredTooltip)}
+  {#if rec}
+    <div
+      class="fixed z-[9999] px-3 py-2 bg-exo-black border border-exo-medium-gray/50 rounded text-xs text-exo-light-gray whitespace-nowrap shadow-lg pointer-events-none"
+      style="left: {tooltipAnchor.x}px; top: {tooltipAnchor.y -
+        8}px; transform: translate(-50%, -100%);"
+    >
+      {rec.tooltip}
+    </div>
+  {/if}
+{/if}

dashboard/src/lib/components/ChatSidebar.svelte

@@ -2,7 +2,6 @@
   import {
     conversations,
     activeConversationId,
-    createConversation,
     loadConversation,
     deleteConversation,
     deleteAllConversations,
@@ -17,9 +16,21 @@
 
   interface Props {
     class?: string;
+    onNewChat?: () => void;
+    onSelectConversation?: () => void;
+    isMobileDrawer?: boolean;
+    isOpen?: boolean;
+    onClose?: () => void;
   }
 
-  let { class: className = "" }: Props = $props();
+  let {
+    class: className = "",
+    onNewChat,
+    onSelectConversation,
+    isMobileDrawer = false,
+    isOpen = false,
+    onClose,
+  }: Props = $props();
 
   const conversationList = $derived(conversations());
   const activeId = $derived(activeConversationId());
@@ -42,11 +53,16 @@
   );
 
   function handleNewChat() {
-    createConversation();
+    onNewChat?.();
   }
 
   function handleSelectConversation(id: string) {
+    onSelectConversation?.();
     loadConversation(id);
+    // Close mobile drawer when selecting a conversation
+    if (isMobileDrawer && isOpen) {
+      onClose?.();
+    }
   }
 
   function handleStartEdit(id: string, name: string, event: MouseEvent) {
@@ -185,11 +201,7 @@
 
     let instanceType: string | null = null;
     if (instanceTag === "MlxRingInstance") instanceType = "MLX Ring";
-    else if (
-      instanceTag === "MlxIbvInstance" ||
-      instanceTag === "MlxJacclInstance"
-    )
-      instanceType = "MLX RDMA";
+    else if (instanceTag === "MlxJacclInstance") instanceType = "MLX RDMA";
 
     let sharding: string | null = null;
     const inst = instance as {
@@ -250,9 +262,7 @@
   }
 </script>
 
-<aside
-  class="flex flex-col h-full bg-exo-dark-gray border-r border-exo-yellow/10 {className}"
->
+{#snippet sidebarContent()}
   <!-- Header -->
   <div class="p-4">
     <button
@@ -307,7 +317,7 @@
       <div class="py-2">
         <div class="px-4 py-2">
           <span
-            class="text-sm text-white/70 font-mono tracking-wider uppercase"
+            class="text-xs text-exo-light-gray font-mono tracking-wider uppercase"
           >
             {searchQuery ? "SEARCH RESULTS" : "CONVERSATIONS"}
           </span>
@@ -376,39 +386,37 @@
                 onkeydown={(e) =>
                   e.key === "Enter" &&
                   handleSelectConversation(conversation.id)}
-                class="group w-full flex items-center justify-between p-2 rounded mb-1 transition-all text-left cursor-pointer
+                class="group w-full flex items-center justify-between p-2.5 rounded-lg mb-1 transition-all text-left cursor-pointer
 									{activeId === conversation.id
-                  ? 'bg-transparent border border-exo-yellow/30'
-                  : 'hover:border-exo-yellow/20 border border-transparent'}"
+                  ? 'bg-exo-yellow/5 border border-exo-yellow/30'
+                  : 'hover:bg-white/[0.03] hover:border-white/10 border border-transparent'}"
               >
                 <div class="flex-1 min-w-0 pr-2">
                   <div
-                    class="text-sm truncate {activeId === conversation.id
+                    class="text-sm font-medium truncate {activeId ===
+                    conversation.id
                       ? 'text-exo-yellow'
-                      : 'text-white/90'}"
+                      : 'text-white'}"
                   >
                     {conversation.name}
                   </div>
-                  <div class="text-sm text-white/50 mt-0.5">
+                  <div class="text-xs text-white/60 mt-0.5">
                     {formatDate(conversation.updatedAt)}
                   </div>
-                  <div class="text-sm text-white/70 truncate">
+                  <div class="text-xs text-exo-light-gray truncate">
                     {info.modelLabel}
                   </div>
-                  <div class="text-xs text-white/60 font-mono">
-                    Strategy: <span class="text-white/80"
-                      >{info.strategyLabel}</span
-                    >
-                  </div>
                   {#if stats}
-                    <div class="text-xs text-white/60 font-mono mt-1">
-                      {#if stats.ttftMs}<span class="text-white/40">TTFT</span>
-                        {stats.ttftMs.toFixed(
-                          0,
-                        )}ms{/if}{#if stats.ttftMs && stats.tps}<span
-                          class="text-white/30 mx-1.5">•</span
-                        >{/if}{#if stats.tps}{stats.tps.toFixed(1)}
-                        <span class="text-white/40">tok/s</span>{/if}
+                    <div class="text-xs text-white/70 font-mono mt-1">
+                      {#if stats.ttftMs}<span class="text-white/50">TTFT</span>
+                        <span class="text-exo-yellow/80"
+                          >{stats.ttftMs.toFixed(0)}ms</span
+                        >{/if}{#if stats.ttftMs && stats.tps}<span
+                          class="text-white/30 mx-1.5">·</span
+                        >{/if}{#if stats.tps}<span class="text-exo-yellow/80"
+                          >{stats.tps.toFixed(1)}</span
+                        >
+                        <span class="text-white/50">tok/s</span>{/if}
                     </div>
                   {/if}
                 </div>
@@ -591,4 +599,30 @@
       </button>
     </div>
   </div>
-</aside>
+{/snippet}
+
+{#if isMobileDrawer}
+  <!-- Mobile drawer with overlay -->
+  {#if isOpen}
+    <!-- Overlay backdrop -->
+    <button
+      type="button"
+      class="fixed inset-0 bg-black/60 backdrop-blur-sm z-40 md:hidden"
+      onclick={() => onClose?.()}
+      aria-label="Close sidebar"
+    ></button>
+    <!-- Drawer panel -->
+    <aside
+      class="fixed left-0 top-0 bottom-0 w-72 bg-exo-dark-gray border-r border-exo-yellow/10 z-50 flex flex-col md:hidden"
+    >
+      {@render sidebarContent()}
+    </aside>
+  {/if}
+{:else}
+  <!-- Desktop sidebar -->
+  <aside
+    class="flex flex-col h-full bg-exo-dark-gray border-r border-exo-yellow/10 {className}"
+  >
+    {@render sidebarContent()}
+  </aside>
+{/if}

dashboard/src/lib/components/ConnectionBanner.svelte

@@ -0,0 +1,20 @@
+<script lang="ts">
+  import { isConnected } from "$lib/stores/app.svelte";
+  import { slide } from "svelte/transition";
+
+  const connected = $derived(isConnected());
+</script>
+
+{#if !connected}
+  <div
+    transition:slide={{ duration: 200 }}
+    class="relative z-50 flex items-center justify-center gap-2 px-4 py-2 bg-red-950/80 border-b border-red-500/30"
+    role="alert"
+    aria-live="assertive"
+  >
+    <div class="w-2 h-2 bg-red-500 rounded-full animate-pulse"></div>
+    <span class="text-xs font-mono text-red-300 tracking-wider uppercase">
+      Connection lost &mdash; Reconnecting to backend&hellip;
+    </span>
+  </div>
+{/if}

dashboard/src/lib/components/DeviceIcon.svelte

@@ -0,0 +1,277 @@
+<script lang="ts">
+  /**
+   * DeviceIcon — renders a device icon as an SVG <g> element.
+   * Uses the exact same proportional math as TopologyGraph.svelte
+   * so that devices look identical in both the topology view and
+   * the onboarding animation.
+   *
+   * Must be placed inside an <svg> element.
+   */
+
+  interface Props {
+    /** "macbook pro" | "mac studio" | "mac mini" etc. */
+    deviceType: string;
+    /** Center X coordinate in SVG space */
+    cx: number;
+    /** Center Y coordinate in SVG space */
+    cy: number;
+    /** Base sizing factor (equivalent to TopologyGraph's nodeRadius) */
+    size?: number;
+    /** RAM usage 0–100 */
+    ramPercent?: number;
+    /** Unique id suffix for clip-path ids */
+    uid?: string;
+  }
+
+  let {
+    deviceType,
+    cx,
+    cy,
+    size = 60,
+    ramPercent = 60,
+    uid = "dev",
+  }: Props = $props();
+
+  // Apple logo path — same constant used by TopologyGraph
+  const APPLE_LOGO_PATH =
+    "M788.1 340.9c-5.8 4.5-108.2 62.2-108.2 190.5 0 148.4 130.3 200.9 134.2 202.2-.6 3.2-20.7 71.9-68.7 141.9-42.8 61.6-87.5 123.1-155.5 123.1s-85.5-39.5-164-39.5c-76.5 0-103.7 40.8-165.9 40.8s-105.6-57-155.5-127C46.7 790.7 0 663 0 541.8c0-194.4 126.4-297.5 250.8-297.5 66.1 0 121.2 43.4 162.7 43.4 39.5 0 101.1-46 176.3-46 28.5 0 130.9 2.6 198.3 99.2zm-234-181.5c31.1-36.9 53.1-88.1 53.1-139.3 0-7.1-.6-14.3-1.9-20.1-50.6 1.9-110.8 33.7-147.1 75.8-28.5 32.4-55.1 83.6-55.1 135.5 0 7.8 1.3 15.6 1.9 18.1 3.2.6 8.4 1.3 13.6 1.3 45.4 0 102.5-30.4 135.5-71.3z";
+  const LOGO_NATIVE_WIDTH = 814;
+  const LOGO_NATIVE_HEIGHT = 1000;
+
+  const wireColor = "rgba(179,179,179,0.8)";
+  const strokeWidth = 1.5;
+
+  const modelLower = $derived(deviceType.toLowerCase());
+
+  // ── Mac Studio dimensions (same ratios as TopologyGraph) ──
+  const studioW = $derived(size * 1.25);
+  const studioH = $derived(size * 0.85);
+  const studioX = $derived(cx - studioW / 2);
+  const studioY = $derived(cy - studioH / 2);
+  const studioCorner = 4;
+  const studioTopH = $derived(studioH * 0.15);
+
+  // Studio front panel details
+  const studioSlotH = $derived(studioH * 0.14);
+  const studioVSlotW = $derived(studioW * 0.05);
+  const studioVSlotY = $derived(
+    studioY + studioTopH + (studioH - studioTopH) * 0.6,
+  );
+  const studioVSlot1X = $derived(studioX + studioW * 0.18);
+  const studioVSlot2X = $derived(studioX + studioW * 0.28);
+  const studioHSlotW = $derived(studioW * 0.2);
+  const studioHSlotX = $derived(studioX + studioW * 0.5 - studioHSlotW / 2);
+
+  // Studio memory fill
+  const studioMemTotalH = $derived(studioH - studioTopH);
+  const studioMemH = $derived((ramPercent / 100) * studioMemTotalH);
+
+  // ── MacBook dimensions (same ratios as TopologyGraph) ──
+  const mbW = $derived((size * 1.6 * 0.85) / 1.15);
+  const mbH = $derived(size * 0.85);
+  const mbX = $derived(cx - mbW / 2);
+  const mbY = $derived(cy - mbH / 2);
+
+  const mbScreenH = $derived(mbH * 0.7);
+  const mbBaseH = $derived(mbH * 0.3);
+  const mbScreenW = $derived(mbW * 0.85);
+  const mbScreenX = $derived(cx - mbScreenW / 2);
+  const mbBezel = 3;
+
+  // MacBook memory fill
+  const mbMemTotalH = $derived(mbScreenH - mbBezel * 2);
+  const mbMemH = $derived((ramPercent / 100) * mbMemTotalH);
+
+  // Apple logo sizing
+  const mbLogoTargetH = $derived(mbScreenH * 0.22);
+  const mbLogoScale = $derived(mbLogoTargetH / LOGO_NATIVE_HEIGHT);
+  const mbLogoX = $derived(cx - (LOGO_NATIVE_WIDTH * mbLogoScale) / 2);
+  const mbLogoY = $derived(
+    mbY + mbScreenH / 2 - (LOGO_NATIVE_HEIGHT * mbLogoScale) / 2,
+  );
+
+  // MacBook base (trapezoidal)
+  const mbBaseY = $derived(mbY + mbScreenH);
+  const mbBaseTopW = $derived(mbScreenW);
+  const mbBaseBottomW = $derived(mbW);
+  const mbBaseTopX = $derived(cx - mbBaseTopW / 2);
+  const mbBaseBottomX = $derived(cx - mbBaseBottomW / 2);
+
+  // Keyboard
+  const mbKbX = $derived(mbBaseTopX + 6);
+  const mbKbY = $derived(mbBaseY + 3);
+  const mbKbW = $derived(mbBaseTopW - 12);
+  const mbKbH = $derived(mbBaseH * 0.55);
+
+  // Trackpad
+  const mbTpW = $derived(mbBaseTopW * 0.4);
+  const mbTpX = $derived(cx - mbTpW / 2);
+  const mbTpY = $derived(mbBaseY + mbKbH + 5);
+  const mbTpH = $derived(mbBaseH * 0.3);
+
+  // Clip IDs
+  const screenClipId = $derived(`di-screen-${uid}`);
+  const studioClipId = $derived(`di-studio-${uid}`);
+</script>
+
+{#if modelLower === "mac studio" || modelLower === "mac mini"}
+  <!-- Mac Studio / Mac Mini -->
+  <defs>
+    <clipPath id={studioClipId}>
+      <rect
+        x={studioX}
+        y={studioY + studioTopH}
+        width={studioW}
+        height={studioH - studioTopH}
+        rx={studioCorner - 1}
+      />
+    </clipPath>
+  </defs>
+
+  <!-- Main body -->
+  <rect
+    x={studioX}
+    y={studioY}
+    width={studioW}
+    height={studioH}
+    rx={studioCorner}
+    fill="#1a1a1a"
+    stroke={wireColor}
+    stroke-width={strokeWidth}
+  />
+
+  <!-- Memory fill -->
+  {#if ramPercent > 0}
+    <rect
+      x={studioX}
+      y={studioY + studioTopH + (studioMemTotalH - studioMemH)}
+      width={studioW}
+      height={studioMemH}
+      fill="rgba(255,215,0,0.75)"
+      clip-path="url(#{studioClipId})"
+    />
+  {/if}
+
+  <!-- Top surface divider -->
+  <line
+    x1={studioX}
+    y1={studioY + studioTopH}
+    x2={studioX + studioW}
+    y2={studioY + studioTopH}
+    stroke="rgba(179,179,179,0.3)"
+    stroke-width="0.5"
+  />
+
+  <!-- Front panel: vertical slots -->
+  <rect
+    x={studioVSlot1X - studioVSlotW / 2}
+    y={studioVSlotY}
+    width={studioVSlotW}
+    height={studioSlotH}
+    fill="rgba(0,0,0,0.35)"
+    rx="1.5"
+  />
+  <rect
+    x={studioVSlot2X - studioVSlotW / 2}
+    y={studioVSlotY}
+    width={studioVSlotW}
+    height={studioSlotH}
+    fill="rgba(0,0,0,0.35)"
+    rx="1.5"
+  />
+
+  <!-- Horizontal slot (SD card) -->
+  <rect
+    x={studioHSlotX}
+    y={studioVSlotY}
+    width={studioHSlotW}
+    height={studioSlotH * 0.6}
+    fill="rgba(0,0,0,0.35)"
+    rx="1"
+  />
+{:else}
+  <!-- MacBook Pro -->
+  <defs>
+    <clipPath id={screenClipId}>
+      <rect
+        x={mbScreenX + mbBezel}
+        y={mbY + mbBezel}
+        width={mbScreenW - mbBezel * 2}
+        height={mbScreenH - mbBezel * 2}
+        rx="2"
+      />
+    </clipPath>
+  </defs>
+
+  <!-- Screen outer frame -->
+  <rect
+    x={mbScreenX}
+    y={mbY}
+    width={mbScreenW}
+    height={mbScreenH}
+    rx="3"
+    fill="#1a1a1a"
+    stroke={wireColor}
+    stroke-width={strokeWidth}
+  />
+
+  <!-- Screen inner (dark) -->
+  <rect
+    x={mbScreenX + mbBezel}
+    y={mbY + mbBezel}
+    width={mbScreenW - mbBezel * 2}
+    height={mbScreenH - mbBezel * 2}
+    rx="2"
+    fill="#0a0a12"
+  />
+
+  <!-- Memory fill on screen -->
+  {#if ramPercent > 0}
+    <rect
+      x={mbScreenX + mbBezel}
+      y={mbY + mbBezel + (mbMemTotalH - mbMemH)}
+      width={mbScreenW - mbBezel * 2}
+      height={mbMemH}
+      fill="rgba(255,215,0,0.85)"
+      clip-path="url(#{screenClipId})"
+    />
+  {/if}
+
+  <!-- Apple logo -->
+  <path
+    d={APPLE_LOGO_PATH}
+    transform="translate({mbLogoX}, {mbLogoY}) scale({mbLogoScale})"
+    fill="#FFFFFF"
+    opacity="0.9"
+  />
+
+  <!-- Keyboard base (trapezoidal) -->
+  <path
+    d="M {mbBaseTopX} {mbBaseY} L {mbBaseTopX +
+      mbBaseTopW} {mbBaseY} L {mbBaseBottomX + mbBaseBottomW} {mbBaseY +
+      mbBaseH} L {mbBaseBottomX} {mbBaseY + mbBaseH} Z"
+    fill="#2c2c2c"
+    stroke={wireColor}
+    stroke-width="1"
+  />
+
+  <!-- Keyboard area -->
+  <rect
+    x={mbKbX}
+    y={mbKbY}
+    width={mbKbW}
+    height={mbKbH}
+    fill="rgba(0,0,0,0.2)"
+    rx="2"
+  />
+
+  <!-- Trackpad -->
+  <rect
+    x={mbTpX}
+    y={mbTpY}
+    width={mbTpW}
+    height={mbTpH}
+    fill="rgba(255,255,255,0.08)"
+    rx="2"
+  />
+{/if}

dashboard/src/lib/components/FamilyLogos.svelte

@@ -82,6 +82,18 @@
       d="M12.025 1.13c-5.77 0-10.449 4.647-10.449 10.378 0 1.112.178 2.181.503 3.185.064-.222.203-.444.416-.577a.96.96 0 0 1 .524-.15c.293 0 .584.124.84.284.278.173.48.408.71.694.226.282.458.611.684.951v-.014c.017-.324.106-.622.264-.874s.403-.487.762-.543c.3-.047.596.06.787.203s.31.313.4.467c.15.257.212.468.233.542.01.026.653 1.552 1.657 2.54.616.605 1.01 1.223 1.082 1.912.055.537-.096 1.059-.38 1.572.637.121 1.294.187 1.967.187.657 0 1.298-.063 1.921-.178-.287-.517-.44-1.041-.384-1.581.07-.69.465-1.307 1.081-1.913 1.004-.987 1.647-2.513 1.657-2.539.021-.074.083-.285.233-.542.09-.154.208-.323.4-.467a1.08 1.08 0 0 1 .787-.203c.359.056.604.29.762.543s.247.55.265.874v.015c.225-.34.457-.67.683-.952.23-.286.432-.52.71-.694.257-.16.547-.284.84-.285a.97.97 0 0 1 .524.151c.228.143.373.388.43.625l.006.04a10.3 10.3 0 0 0 .534-3.273c0-5.731-4.678-10.378-10.449-10.378M8.327 6.583a1.5 1.5 0 0 1 .713.174 1.487 1.487 0 0 1 .617 2.013c-.183.343-.762-.214-1.102-.094-.38.134-.532.914-.917.71a1.487 1.487 0 0 1 .69-2.803m7.486 0a1.487 1.487 0 0 1 .689 2.803c-.385.204-.536-.576-.916-.71-.34-.12-.92.437-1.103.094a1.487 1.487 0 0 1 .617-2.013 1.5 1.5 0 0 1 .713-.174m-10.68 1.55a.96.96 0 1 1 0 1.921.96.96 0 0 1 0-1.92m13.838 0a.96.96 0 1 1 0 1.92.96.96 0 0 1 0-1.92M8.489 11.458c.588.01 1.965 1.157 3.572 1.164 1.607-.007 2.984-1.155 3.572-1.164.196-.003.305.12.305.454 0 .886-.424 2.328-1.563 3.202-.22-.756-1.396-1.366-1.63-1.32q-.011.001-.02.006l-.044.026-.01.008-.03.024q-.018.017-.035.036l-.032.04a1 1 0 0 0-.058.09l-.014.025q-.049.088-.11.19a1 1 0 0 1-.083.116 1.2 1.2 0 0 1-.173.18q-.035.029-.075.058a1.3 1.3 0 0 1-.251-.243 1 1 0 0 1-.076-.107c-.124-.193-.177-.363-.337-.444-.034-.016-.104-.008-.2.022q-.094.03-.216.087-.06.028-.125.063l-.13.074q-.067.04-.136.086a3 3 0 0 0-.135.096 3 3 0 0 0-.26.219 2 2 0 0 0-.12.121 2 2 0 0 0-.106.128l-.002.002a2 2 0 0 0-.09.132l-.001.001a1.2 1.2 0 0 0-.105.212q-.013.036-.024.073c-1.139-.875-1.563-2.317-1.563-3.203 0-.334.109-.457.305-.454m.836 10.354c.824-1.19.766-2.082-.365-3.194-1.13-1.112-1.789-2.738-1.789-2.738s-.246-.945-.806-.858-.97 1.499.202 2.362c1.173.864-.233 1.45-.685.64-.45-.812-1.683-2.896-2.322-3.295s-1.089-.175-.938.647 2.822 2.813 2.562 3.244-1.176-.506-1.176-.506-2.866-2.567-3.49-1.898.473 1.23 2.037 2.16c1.564.932 1.686 1.178 1.464 1.53s-3.675-2.511-4-1.297c-.323 1.214 3.524 1.567 3.287 2.405-.238.839-2.71-1.587-3.216-.642-.506.946 3.49 2.056 3.522 2.064 1.29.33 4.568 1.028 5.713-.624m5.349 0c-.824-1.19-.766-2.082.365-3.194 1.13-1.112 1.789-2.738 1.789-2.738s.246-.945.806-.858.97 1.499-.202 2.362c-1.173.864.233 1.45.685.64.451-.812 1.683-2.896 2.322-3.295s1.089-.175.938.647-2.822 2.813-2.562 3.244 1.176-.506 1.176-.506 2.866-2.567 3.49-1.898-.473 1.23-2.037 2.16c-1.564.932-1.686 1.178-1.464 1.53s3.675-2.511 4-1.297c.323 1.214-3.524 1.567-3.287 2.405.238.839 2.71-1.587 3.216-.642.506.946-3.49 2.056-3.522 2.064-1.29.33-4.568 1.028-5.713-.624"
     />
   </svg>
+{:else if family === "step"}
+  <svg class="w-6 h-6 {className}" viewBox="0 0 24 24" fill="currentColor">
+    <path
+      d="M22.012 0h1.032v.927H24v.968h-.956V3.78h-1.032V1.896h-1.878v-.97h1.878V0zM2.6 12.371V1.87h.969v10.502h-.97zm10.423.66h10.95v.918h-6.208v9.579h-4.742V13.03zM5.629 3.333v12.356H0v4.51h10.386V8L20.859 8l-.003-4.668-15.227.001z"
+    />
+  </svg>
+{:else if family === "nemotron"}
+  <svg class="w-6 h-6 {className}" viewBox="0 0 24 24" fill="currentColor">
+    <path
+      d="M8.948 8.798v-1.43a6.7 6.7 0 0 1 .424-.018c3.922-.124 6.493 3.374 6.493 3.374s-2.774 3.851-5.75 3.851c-.398 0-.787-.062-1.158-.185v-4.346c1.528.185 1.837.857 2.747 2.385l2.04-1.714s-1.492-1.952-4-1.952a6.016 6.016 0 0 0-.796.035m0-4.735v2.138l.424-.027c5.45-.185 9.01 4.47 9.01 4.47s-4.08 4.964-8.33 4.964c-.37 0-.733-.035-1.095-.097v1.325c.3.035.61.062.91.062 3.957 0 6.82-2.023 9.593-4.408.459.371 2.34 1.263 2.73 1.652-2.633 2.208-8.772 3.984-12.253 3.984-.335 0-.653-.018-.971-.053v1.864H24V4.063zm0 10.326v1.131c-3.657-.654-4.673-4.46-4.673-4.46s1.758-1.944 4.673-2.262v1.237H8.94c-1.528-.186-2.73 1.245-2.73 1.245s.68 2.412 2.739 3.11M2.456 10.9s2.164-3.197 6.5-3.533V6.201C4.153 6.59 0 10.653 0 10.653s2.35 6.802 8.948 7.42v-1.237c-4.84-.6-6.492-5.936-6.492-5.936z"
+    />
+  </svg>
 {:else}
   <svg class="w-6 h-6 {className}" viewBox="0 0 24 24" fill="currentColor">
     <path

dashboard/src/lib/components/FamilySidebar.svelte

@@ -31,6 +31,7 @@
     kimi: "Kimi",
     flux: "FLUX",
     "qwen-image": "Qwen Img",
+    nemotron: "NVIDIA",
   };
 
   function getFamilyName(family: string): string {
@@ -41,31 +42,20 @@
 </script>
 
 <div
-  class="flex flex-col gap-1 py-2 px-1 border-r border-exo-yellow/10 bg-exo-medium-gray/30 min-w-[64px] overflow-y-auto scrollbar-hide"
+  class="flex flex-col gap-1 py-2 px-1 border-r border-exo-yellow/10 bg-exo-medium-gray/30 min-w-[80px] sm:min-w-[72px] overflow-y-auto scrollbar-hide"
 >
   <!-- All models (no filter) -->
   <button
     type="button"
     onclick={() => onSelect(null)}
-    class="group flex flex-col items-center justify-center p-2 rounded transition-all duration-200 cursor-pointer {selectedFamily ===
+    class="group flex items-center justify-center px-3 py-2.5 rounded transition-all duration-200 cursor-pointer min-h-[44px] sm:min-h-0 {selectedFamily ===
     null
       ? 'bg-exo-yellow/20 border-l-2 border-exo-yellow'
       : 'hover:bg-white/5 border-l-2 border-transparent'}"
     title="All models"
   >
-    <svg
-      class="w-5 h-5 {selectedFamily === null
-        ? 'text-exo-yellow'
-        : 'text-white/50 group-hover:text-white/70'}"
-      viewBox="0 0 24 24"
-      fill="currentColor"
-    >
-      <path
-        d="M4 8h4V4H4v4zm6 12h4v-4h-4v4zm-6 0h4v-4H4v4zm0-6h4v-4H4v4zm6 0h4v-4h-4v4zm6-10v4h4V4h-4zm-6 4h4V4h-4v4zm6 6h4v-4h-4v4zm0 6h4v-4h-4v4z"
-      />
-    </svg>
     <span
-      class="text-[9px] font-mono mt-0.5 {selectedFamily === null
+      class="text-[12px] font-mono font-medium {selectedFamily === null
         ? 'text-exo-yellow'
         : 'text-white/40 group-hover:text-white/60'}">All</span
     >
@@ -89,7 +79,7 @@
           : "text-white/50 group-hover:text-amber-400/70"}
       />
       <span
-        class="text-[9px] font-mono mt-0.5 {selectedFamily === 'favorites'
+        class="text-[11px] font-mono mt-0.5 {selectedFamily === 'favorites'
           ? 'text-amber-400'
           : 'text-white/40 group-hover:text-white/60'}">Faves</span
       >
@@ -114,7 +104,7 @@
           : "text-white/50 group-hover:text-white/70"}
       />
       <span
-        class="text-[9px] font-mono mt-0.5 {selectedFamily === 'recents'
+        class="text-[11px] font-mono mt-0.5 {selectedFamily === 'recents'
           ? 'text-exo-yellow'
           : 'text-white/40 group-hover:text-white/60'}">Recent</span
       >
@@ -138,7 +128,7 @@
         : "text-white/50 group-hover:text-orange-400/70"}
     />
     <span
-      class="text-[9px] font-mono mt-0.5 {selectedFamily === 'huggingface'
+      class="text-[11px] font-mono mt-0.5 {selectedFamily === 'huggingface'
         ? 'text-orange-400'
         : 'text-white/40 group-hover:text-white/60'}">Hub</span
     >
@@ -164,7 +154,7 @@
           : "text-white/50 group-hover:text-white/70"}
       />
       <span
-        class="text-[9px] font-mono mt-0.5 truncate max-w-full {selectedFamily ===
+        class="text-[11px] font-mono mt-0.5 truncate max-w-full {selectedFamily ===
         family
           ? 'text-exo-yellow'
           : 'text-white/40 group-hover:text-white/60'}"

dashboard/src/lib/components/HeaderNav.svelte

@@ -6,6 +6,16 @@
   export let showSidebarToggle = false;
   export let sidebarVisible = true;
   export let onToggleSidebar: (() => void) | null = null;
+  export let showMobileMenuToggle = false;
+  export let mobileMenuOpen = false;
+  export let onToggleMobileMenu: (() => void) | null = null;
+  export let showMobileRightToggle = false;
+  export let mobileRightOpen = false;
+  export let onToggleMobileRight: (() => void) | null = null;
+  export let downloadProgress: {
+    count: number;
+    percentage: number;
+  } | null = null;
 
   function handleHome(): void {
     if (onHome) {
@@ -23,45 +33,96 @@
       onToggleSidebar();
     }
   }
+
+  function handleToggleMobileMenu(): void {
+    if (onToggleMobileMenu) {
+      onToggleMobileMenu();
+    }
+  }
+
+  function handleToggleMobileRight(): void {
+    if (onToggleMobileRight) {
+      onToggleMobileRight();
+    }
+  }
 </script>
 
 <header
-  class="relative z-20 flex items-center justify-center px-6 pt-8 pb-4 bg-exo-dark-gray"
+  class="relative z-20 flex items-center justify-center px-4 md:px-6 pt-4 md:pt-8 pb-3 md:pb-4 bg-exo-dark-gray"
 >
-  <!-- Left: Sidebar Toggle -->
-  {#if showSidebarToggle}
-    <div class="absolute left-6 top-1/2 -translate-y-1/2">
-      <button
-        onclick={handleToggleSidebar}
-        class="p-2 rounded border border-exo-medium-gray/40 hover:border-exo-yellow/50 transition-colors cursor-pointer"
-        title={sidebarVisible ? "Hide sidebar" : "Show sidebar"}
+  <!-- Left: Sidebar Toggle (desktop) or Mobile Sidebar Toggle (mobile) -->
+  <div
+    class="absolute left-4 md:left-6 top-1/2 -translate-y-1/2 flex items-center gap-2"
+  >
+    <!-- Mobile sidebar toggle -->
+    <button
+      onclick={handleToggleMobileMenu}
+      class="p-2 rounded border border-exo-light-gray/30 hover:border-exo-yellow/50 hover:bg-exo-medium-gray/30 transition-colors cursor-pointer md:hidden"
+      title={mobileMenuOpen ? "Hide sidebar" : "Show sidebar"}
+      aria-label={mobileMenuOpen
+        ? "Hide conversation sidebar"
+        : "Show conversation sidebar"}
+      aria-pressed={mobileMenuOpen}
+    >
+      <svg
+        fill="none"
+        viewBox="0 0 24 24"
+        stroke="currentColor"
+        stroke-width="2"
+        class="w-5 h-5 {mobileMenuOpen
+          ? 'text-exo-yellow'
+          : 'text-exo-light-gray'}"
       >
-        <svg
-          class="w-5 h-5 {sidebarVisible
-            ? 'text-exo-yellow'
-            : 'text-exo-medium-gray'}"
-          fill="none"
-          viewBox="0 0 24 24"
-          stroke="currentColor"
-          stroke-width="2"
-        >
-          {#if sidebarVisible}
-            <path
-              stroke-linecap="round"
-              stroke-linejoin="round"
-              d="M11 19l-7-7 7-7m8 14l-7-7 7-7"
-            />
-          {:else}
-            <path
-              stroke-linecap="round"
-              stroke-linejoin="round"
-              d="M13 5l7 7-7 7M5 5l7 7-7 7"
-            />
-          {/if}
-        </svg>
-      </button>
-    </div>
-  {/if}
+        {#if mobileMenuOpen}
+          <path
+            stroke-linecap="round"
+            stroke-linejoin="round"
+            d="M11 19l-7-7 7-7m8 14l-7-7 7-7"
+          ></path>
+        {:else}
+          <path
+            stroke-linecap="round"
+            stroke-linejoin="round"
+            d="M13 5l7 7-7 7M5 5l7 7-7 7"
+          ></path>
+        {/if}
+      </svg>
+    </button>
+    <!-- Desktop sidebar toggle -->
+    <button
+      onclick={handleToggleSidebar}
+      class="p-2 rounded border border-exo-light-gray/30 hover:border-exo-yellow/50 hover:bg-exo-medium-gray/30 transition-colors cursor-pointer hidden md:block"
+      title={sidebarVisible ? "Hide sidebar" : "Show sidebar"}
+      aria-label={sidebarVisible
+        ? "Hide conversation sidebar"
+        : "Show conversation sidebar"}
+      aria-pressed={sidebarVisible}
+    >
+      <svg
+        fill="none"
+        viewBox="0 0 24 24"
+        stroke="currentColor"
+        stroke-width="2"
+        class="w-5 h-5 {sidebarVisible
+          ? 'text-exo-yellow'
+          : 'text-exo-light-gray'}"
+      >
+        {#if sidebarVisible}
+          <path
+            stroke-linecap="round"
+            stroke-linejoin="round"
+            d="M11 19l-7-7 7-7m8 14l-7-7 7-7"
+          ></path>
+        {:else}
+          <path
+            stroke-linecap="round"
+            stroke-linejoin="round"
+            d="M13 5l7 7-7 7M5 5l7 7-7 7"
+          ></path>
+        {/if}
+      </svg>
+    </button>
+  </div>
 
   <!-- Center: Logo (clickable to go home) -->
   <button
@@ -75,18 +136,55 @@
     <img
       src="/exo-logo.png"
       alt="EXO"
-      class="h-18 drop-shadow-[0_0_20px_rgba(255,215,0,0.5)]"
+      class="h-12 md:h-18 drop-shadow-[0_0_4px_rgba(255,215,0,0.3)]"
     />
   </button>
 
-  <!-- Right: Home + Downloads -->
-  <div
-    class="absolute right-6 top-1/2 -translate-y-1/2 flex items-center gap-4"
+  <!-- Right: Home + Downloads + Mobile Right Toggle -->
+  <nav
+    class="absolute right-4 md:right-6 top-1/2 -translate-y-1/2 flex items-center gap-2 md:gap-4"
+    aria-label="Main navigation"
   >
+    <!-- Mobile right sidebar toggle (instances/models) - only show when not in chat mode -->
+    {#if showMobileRightToggle}
+      <button
+        onclick={handleToggleMobileRight}
+        class="p-2 rounded border border-exo-light-gray/30 hover:border-exo-yellow/50 hover:bg-exo-medium-gray/30 transition-colors cursor-pointer md:hidden"
+        title={mobileRightOpen ? "Hide instances" : "Show instances"}
+        aria-label={mobileRightOpen
+          ? "Hide instances panel"
+          : "Show instances panel"}
+        aria-pressed={mobileRightOpen}
+      >
+        <svg
+          fill="none"
+          viewBox="0 0 24 24"
+          stroke="currentColor"
+          stroke-width="2"
+          class="w-5 h-5 {mobileRightOpen
+            ? 'text-exo-yellow'
+            : 'text-exo-light-gray'}"
+        >
+          {#if mobileRightOpen}
+            <path
+              stroke-linecap="round"
+              stroke-linejoin="round"
+              d="M13 5l7 7-7 7M5 5l7 7-7 7"
+            ></path>
+          {:else}
+            <path
+              stroke-linecap="round"
+              stroke-linejoin="round"
+              d="M11 19l-7-7 7-7m8 14l-7-7 7-7"
+            ></path>
+          {/if}
+        </svg>
+      </button>
+    {/if}
     {#if showHome}
       <button
         onclick={handleHome}
-        class="text-sm text-exo-light-gray hover:text-exo-yellow transition-colors tracking-wider uppercase flex items-center gap-2 cursor-pointer"
+        class="flex text-sm text-white/70 hover:text-exo-yellow transition-colors tracking-wider uppercase items-center gap-2 cursor-pointer"
         title="Back to topology view"
       >
         <svg
@@ -102,28 +200,64 @@
             d="M3 12l2-2m0 0l7-7 7 7M5 10v10a1 1 0 001 1h3m10-11l2 2m-2-2v10a1 1 0 01-1 1h-3m-6 0a1 1 0 001-1v-4a1 1 0 011-1h2a1 1 0 011 1v4a1 1 0 001 1m-6 0h6"
           />
         </svg>
-        Home
+        <span class="hidden sm:inline">Home</span>
       </button>
     {/if}
     <a
       href="/#/downloads"
-      class="text-sm text-exo-light-gray hover:text-exo-yellow transition-colors tracking-wider uppercase flex items-center gap-2 cursor-pointer"
+      class="text-xs md:text-sm text-white/70 hover:text-exo-yellow transition-colors tracking-wider uppercase flex items-center gap-1.5 md:gap-2 cursor-pointer"
       title="View downloads overview"
     >
-      <svg
-        class="w-4 h-4"
-        viewBox="0 0 24 24"
-        fill="none"
-        stroke="currentColor"
-        stroke-width="2"
-        stroke-linecap="round"
-        stroke-linejoin="round"
-      >
-        <path d="M12 3v12" />
-        <path d="M7 12l5 5 5-5" />
-        <path d="M5 21h14" />
-      </svg>
-      Downloads
+      {#if downloadProgress}
+        <!-- Compact download progress indicator -->
+        <div class="relative w-4 h-4 flex-shrink-0">
+          <svg class="w-4 h-4 -rotate-90" viewBox="0 0 20 20">
+            <circle
+              cx="10"
+              cy="10"
+              r="8"
+              fill="none"
+              stroke="currentColor"
+              stroke-width="2"
+              opacity="0.2"
+            />
+            <circle
+              cx="10"
+              cy="10"
+              r="8"
+              fill="none"
+              stroke="currentColor"
+              stroke-width="2"
+              stroke-dasharray={2 * Math.PI * 8}
+              stroke-dashoffset={2 *
+                Math.PI *
+                8 *
+                (1 - downloadProgress.percentage / 100)}
+              class="text-blue-400 transition-all duration-300"
+            />
+          </svg>
+          <div
+            class="absolute inset-0 flex items-center justify-center text-[6px] font-mono text-blue-400"
+          >
+            {downloadProgress.count}
+          </div>
+        </div>
+      {:else}
+        <svg
+          class="w-4 h-4"
+          viewBox="0 0 24 24"
+          fill="none"
+          stroke="currentColor"
+          stroke-width="2"
+          stroke-linecap="round"
+          stroke-linejoin="round"
+        >
+          <path d="M12 3v12" />
+          <path d="M7 12l5 5 5-5" />
+          <path d="M5 21h14" />
+        </svg>
+      {/if}
+      <span class="hidden sm:inline">Downloads</span>
     </a>
-  </div>
+  </nav>
 </header>

dashboard/src/lib/components/HuggingFaceResultItem.svelte

@@ -26,7 +26,8 @@
     downloadedOnNodes = [],
   }: HuggingFaceResultItemProps = $props();
 
-  function formatNumber(num: number): string {
+  function formatNumber(num: number | undefined): string {
+    if (num == null) return "0";
     if (num >= 1000000) {
       return `${(num / 1000000).toFixed(1)}M`;
     } else if (num >= 1000) {
@@ -35,8 +36,12 @@
     return num.toString();
   }
 
-  // Extract model name from full ID (e.g., "mlx-community/Llama-3.2-1B" -> "Llama-3.2-1B")
-  const modelName = $derived(model.id.split("/").pop() || model.id);
+  // Show short name for mlx-community models, full ID for everything else
+  const modelName = $derived(
+    model.author === "mlx-community"
+      ? model.id.split("/").pop() || model.id
+      : model.id,
+  );
 </script>
 
 <div

dashboard/src/lib/components/ImageParamsPanel.svelte

@@ -59,13 +59,14 @@
   }
 
   const sizeOptions: ImageGenerationParams["size"][] = [
+    "auto",
     "512x512",
     "768x768",
     "1024x1024",
     "1024x768",
     "768x1024",
-    "1024x1365",
-    "1365x1024",
+    "1024x1536",
+    "1536x1024",
   ];
 
   const qualityOptions: ImageGenerationParams["quality"][] = [
@@ -176,92 +177,90 @@
 <div class="border-b border-exo-medium-gray/30 px-3 py-2">
   <!-- Basic params row -->
   <div class="flex items-center gap-3 flex-wrap">
-    <!-- Size (hidden in edit mode - output size comes from input image) -->
-    {#if !isEditMode}
-      <div class="flex items-center gap-1.5">
-        <span class="text-xs text-exo-light-gray uppercase tracking-wider"
-          >SIZE:</span
+    <!-- Size -->
+    <div class="flex items-center gap-1.5">
+      <span class="text-xs text-exo-light-gray uppercase tracking-wider"
+        >SIZE:</span
+      >
+      <div class="relative">
+        <button
+          bind:this={sizeButtonRef}
+          type="button"
+          onclick={() => (isSizeDropdownOpen = !isSizeDropdownOpen)}
+          class="bg-exo-medium-gray/50 border border-exo-yellow/30 rounded pl-2 pr-6 py-1 text-xs font-mono text-exo-yellow cursor-pointer transition-all duration-200 hover:border-exo-yellow/50 focus:outline-none focus:border-exo-yellow/70 {isSizeDropdownOpen
+            ? 'border-exo-yellow/70'
+            : ''}"
         >
-        <div class="relative">
-          <button
-            bind:this={sizeButtonRef}
-            type="button"
-            onclick={() => (isSizeDropdownOpen = !isSizeDropdownOpen)}
-            class="bg-exo-medium-gray/50 border border-exo-yellow/30 rounded pl-2 pr-6 py-1 text-xs font-mono text-exo-yellow cursor-pointer transition-all duration-200 hover:border-exo-yellow/50 focus:outline-none focus:border-exo-yellow/70 {isSizeDropdownOpen
-              ? 'border-exo-yellow/70'
-              : ''}"
+          {params.size.toUpperCase()}
+        </button>
+        <div
+          class="absolute right-1.5 top-1/2 -translate-y-1/2 pointer-events-none transition-transform duration-200 {isSizeDropdownOpen
+            ? 'rotate-180'
+            : ''}"
+        >
+          <svg
+            class="w-3 h-3 text-exo-yellow/60"
+            fill="none"
+            viewBox="0 0 24 24"
+            stroke="currentColor"
           >
-            {params.size}
-          </button>
-          <div
-            class="absolute right-1.5 top-1/2 -translate-y-1/2 pointer-events-none transition-transform duration-200 {isSizeDropdownOpen
-              ? 'rotate-180'
-              : ''}"
-          >
-            <svg
-              class="w-3 h-3 text-exo-yellow/60"
-              fill="none"
-              viewBox="0 0 24 24"
-              stroke="currentColor"
-            >
-              <path
-                stroke-linecap="round"
-                stroke-linejoin="round"
-                stroke-width="2"
-                d="M19 9l-7 7-7-7"
-              />
-            </svg>
+            <path
+              stroke-linecap="round"
+              stroke-linejoin="round"
+              stroke-width="2"
+              d="M19 9l-7 7-7-7"
+            />
+          </svg>
+        </div>
+      </div>
+
+      {#if isSizeDropdownOpen}
+        <!-- Backdrop to close dropdown -->
+        <button
+          type="button"
+          class="fixed inset-0 z-[9998] cursor-default"
+          onclick={() => (isSizeDropdownOpen = false)}
+          aria-label="Close dropdown"
+        ></button>
+
+        <!-- Dropdown Panel - fixed positioning to escape overflow:hidden -->
+        <div
+          class="fixed bg-exo-dark-gray border border-exo-yellow/30 rounded shadow-lg shadow-black/50 z-[9999] max-h-48 overflow-y-auto overflow-x-hidden min-w-max"
+          style="bottom: calc(100vh - {sizeDropdownPosition()
+            .top}px + 4px); left: {sizeDropdownPosition().left}px;"
+        >
+          <div class="py-1">
+            {#each sizeOptions as size}
+              <button
+                type="button"
+                onclick={() => selectSize(size)}
+                class="w-full px-3 py-1.5 text-left text-xs font-mono tracking-wide transition-colors duration-100 flex items-center gap-2 {params.size ===
+                size
+                  ? 'bg-transparent text-exo-yellow'
+                  : 'text-exo-light-gray hover:text-exo-yellow'}"
+              >
+                {#if params.size === size}
+                  <svg
+                    class="w-3 h-3 flex-shrink-0"
+                    fill="currentColor"
+                    viewBox="0 0 20 20"
+                  >
+                    <path
+                      fill-rule="evenodd"
+                      d="M16.707 5.293a1 1 0 010 1.414l-8 8a1 1 0 01-1.414 0l-4-4a1 1 0 011.414-1.414L8 12.586l7.293-7.293a1 1 0 011.414 0z"
+                      clip-rule="evenodd"
+                    />
+                  </svg>
+                {:else}
+                  <span class="w-3"></span>
+                {/if}
+                <span>{size.toUpperCase()}</span>
+              </button>
+            {/each}
           </div>
         </div>
-
-        {#if isSizeDropdownOpen}
-          <!-- Backdrop to close dropdown -->
-          <button
-            type="button"
-            class="fixed inset-0 z-[9998] cursor-default"
-            onclick={() => (isSizeDropdownOpen = false)}
-            aria-label="Close dropdown"
-          ></button>
-
-          <!-- Dropdown Panel - fixed positioning to escape overflow:hidden -->
-          <div
-            class="fixed bg-exo-dark-gray border border-exo-yellow/30 rounded shadow-lg shadow-black/50 z-[9999] max-h-48 overflow-y-auto min-w-max"
-            style="bottom: calc(100vh - {sizeDropdownPosition()
-              .top}px + 4px); left: {sizeDropdownPosition().left}px;"
-          >
-            <div class="py-1">
-              {#each sizeOptions as size}
-                <button
-                  type="button"
-                  onclick={() => selectSize(size)}
-                  class="w-full px-3 py-1.5 text-left text-xs font-mono tracking-wide transition-colors duration-100 flex items-center gap-2 {params.size ===
-                  size
-                    ? 'bg-transparent text-exo-yellow'
-                    : 'text-exo-light-gray hover:text-exo-yellow'}"
-                >
-                  {#if params.size === size}
-                    <svg
-                      class="w-3 h-3 flex-shrink-0"
-                      fill="currentColor"
-                      viewBox="0 0 20 20"
-                    >
-                      <path
-                        fill-rule="evenodd"
-                        d="M16.707 5.293a1 1 0 010 1.414l-8 8a1 1 0 01-1.414 0l-4-4a1 1 0 011.414-1.414L8 12.586l7.293-7.293a1 1 0 011.414 0z"
-                        clip-rule="evenodd"
-                      />
-                    </svg>
-                  {:else}
-                    <span class="w-3"></span>
-                  {/if}
-                  <span>{size}</span>
-                </button>
-              {/each}
-            </div>
-          </div>
-        {/if}
-      </div>
-    {/if}
+      {/if}
+    </div>
 
     <!-- Quality -->
     <div class="flex items-center gap-1.5">
@@ -311,7 +310,7 @@
 
         <!-- Dropdown Panel - fixed positioning to escape overflow:hidden -->
         <div
-          class="fixed bg-exo-dark-gray border border-exo-yellow/30 rounded shadow-lg shadow-black/50 z-[9999] max-h-48 overflow-y-auto min-w-max"
+          class="fixed bg-exo-dark-gray border border-exo-yellow/30 rounded shadow-lg shadow-black/50 z-[9999] max-h-48 overflow-y-auto overflow-x-hidden min-w-max"
           style="bottom: calc(100vh - {qualityDropdownPosition()
             .top}px + 4px); left: {qualityDropdownPosition().left}px;"
         >

dashboard/src/lib/components/MarkdownContent.svelte

@@ -507,9 +507,29 @@
   });
 
   $effect(() => {
-    if (containerRef && processedHtml) {
-      setupCopyButtons();
+    if (!containerRef || !browser) return;
+
+    function handleDelegatedClick(event: MouseEvent) {
+      const codeBtn = (event.target as HTMLElement).closest(
+        ".copy-code-btn",
+      ) as HTMLButtonElement | null;
+      if (codeBtn) {
+        handleCopyClick({ currentTarget: codeBtn } as unknown as Event);
+        return;
+      }
+      const mathBtn = (event.target as HTMLElement).closest(
+        ".copy-math-btn",
+      ) as HTMLButtonElement | null;
+      if (mathBtn) {
+        handleMathCopyClick({ currentTarget: mathBtn } as unknown as Event);
+        return;
+      }
     }
+
+    containerRef.addEventListener("click", handleDelegatedClick);
+    return () => {
+      containerRef?.removeEventListener("click", handleDelegatedClick);
+    };
   });
 </script>
 

dashboard/src/lib/components/ModelCard.svelte

@@ -16,12 +16,14 @@
       perNode?: Array<{
         nodeId: string;
         nodeName: string;
-        progress: DownloadProgress;
+        status: "completed" | "partial" | "pending" | "downloading";
+        percentage: number;
+        progress: DownloadProgress | null;
       }>;
     } | null;
     nodes?: Record<string, NodeInfo>;
     sharding?: "Pipeline" | "Tensor";
-    runtime?: "MlxRing" | "MlxIbv" | "MlxJaccl";
+    runtime?: "MlxRing" | "MlxJaccl";
     onLaunch?: () => void;
     tags?: string[];
     apiPreview?: PlacementPreview | null;
@@ -145,10 +147,7 @@
     return `${s}s`;
   }
 
-  const isDownloading = $derived(downloadStatus?.isDownloading ?? false);
-  const progress = $derived(downloadStatus?.progress);
-  const percentage = $derived(progress?.percentage ?? 0);
-  let expandedNodes = $state<Set<string>>(new Set());
+  const perNode = $derived(downloadStatus?.perNode ?? []);
 
   function toggleNodeDetails(nodeId: string): void {
     const next = new Set(expandedNodes);
@@ -348,7 +347,7 @@
   // Debug mode state
   const isDebugMode = $derived(debugMode());
   const topology = $derived(topologyData());
-  const isRdma = $derived(runtime === "MlxIbv" || runtime === "MlxJaccl");
+  const isRdma = $derived(runtime === "MlxJaccl");
 
   // Get interface name for an IP from node data
   function getInterfaceForIp(nodeId: string, ip?: string): string | null {
@@ -567,20 +566,72 @@
     <div class="flex items-center gap-1.5 mb-2">
       <span
         class="px-1.5 py-0.5 text-xs font-mono tracking-wider uppercase bg-exo-medium-gray/30 text-exo-light-gray border border-exo-medium-gray/40"
+        title={sharding === "Pipeline"
+          ? "Pipeline: splits model into sequential stages across devices. Lower network overhead."
+          : "Tensor: splits each layer across devices. Best with high-bandwidth connections (Thunderbolt)."}
       >
         {sharding}
       </span>
       <span
         class="px-1.5 py-0.5 text-xs font-mono tracking-wider uppercase bg-exo-medium-gray/30 text-exo-light-gray border border-exo-medium-gray/40"
+        title={runtime === "MlxRing"
+          ? "Ring: standard networking. Works over any connection (Wi-Fi, Ethernet, Thunderbolt)."
+          : "RDMA: direct memory access over Thunderbolt. Significantly faster for multi-device inference."}
       >
         {runtime === "MlxRing"
           ? "MLX Ring"
-          : runtime === "MlxIbv" || runtime === "MlxJaccl"
+          : runtime === "MlxJaccl"
             ? "MLX RDMA"
             : runtime}
       </span>
     </div>
 
+    <!-- Download Status (per-node) -->
+    {#if perNode.length > 0}
+      <div class="mb-2 space-y-1">
+        <div
+          class="text-[10px] font-mono text-white/20 tracking-widest uppercase"
+        >
+          Download progress
+        </div>
+        {#each perNode as node}
+          <div class="flex items-center gap-2 text-xs font-mono">
+            <span class="text-white/40 w-20 truncate" title={node.nodeId}
+              >{node.nodeName}</span
+            >
+            <div
+              class="flex-1 h-1 bg-exo-medium-gray/30 rounded overflow-hidden"
+            >
+              <div
+                class="h-full transition-all duration-300 {node.status ===
+                'downloading'
+                  ? 'bg-blue-500/70'
+                  : node.status === 'completed'
+                    ? 'bg-exo-yellow/40'
+                    : 'bg-white/20'}"
+                style="width: {node.percentage}%"
+              ></div>
+            </div>
+            <span
+              class="text-right {node.status === 'completed'
+                ? 'text-exo-yellow/60'
+                : node.status === 'downloading'
+                  ? 'text-blue-400/60'
+                  : 'text-white/30'}"
+            >
+              {#if node.status === "downloading" && node.progress}
+                {Math.round(node.percentage)}% {formatSpeed(
+                  node.progress.speed,
+                )}
+              {:else}
+                {node.percentage > 0 ? `${Math.round(node.percentage)}%` : "0%"}
+              {/if}
+            </span>
+          </div>
+        {/each}
+      </div>
+    {/if}
+
     <!-- Mini Topology Preview -->
     {#if placementPreview().nodes.length > 0}
       {@const preview = placementPreview()}
@@ -636,15 +687,7 @@
             {@const allConnections =
               isDebugMode && usedNodes.length > 1
                 ? (() => {
-                    const conns: Array<{
-                      ip: string;
-                      iface: string | null;
-                      from: string;
-                      to: string;
-                      midX: number;
-                      midY: number;
-                      arrow: string;
-                    }> = [];
+                    const conns: Array = [];
                     for (let i = 0; i < usedNodes.length; i++) {
                       for (let j = i + 1; j < usedNodes.length; j++) {
                         const n1 = usedNodes[i];
@@ -656,7 +699,12 @@
                           const toPos = nodePositions[c.to];
                           const arrow =
                             fromPos && toPos ? getArrow(fromPos, toPos) : "→";
-                          conns.push({ ...c, midX, midY, arrow });
+                          conns.push({
+                            ...c,
+                            midX,
+                            midY,
+                            arrow,
+                          });
                         }
                       }
                     }

dashboard/src/lib/components/ModelFilterPopover.svelte

@@ -6,6 +6,7 @@
     capabilities: string[];
     sizeRange: { min: number; max: number } | null;
     downloadedOnly: boolean;
+    readyOnly: boolean;
   }
 
   type ModelFilterPopoverProps = {
@@ -72,7 +73,7 @@
 
 <!-- svelte-ignore a11y_no_static_element_interactions -->
 <div
-  class="filter-popover absolute right-0 top-full mt-2 w-64 bg-exo-dark-gray border border-exo-yellow/10 rounded-lg shadow-xl z-10"
+  class="filter-popover absolute right-0 top-full mt-2 w-64 bg-exo-dark-gray border border-exo-yellow/10 rounded-lg shadow-xl z-20"
   transition:fly={{ y: -10, duration: 200, easing: cubicOut }}
   onclick={(e) => e.stopPropagation()}
   role="dialog"
@@ -190,34 +191,58 @@
       </div>
     </div>
 
-    <!-- Downloaded only -->
+    <!-- Availability filters -->
     <div>
       <h4 class="text-xs font-mono text-white/50 mb-2">Availability</h4>
-      <button
-        type="button"
-        class="px-2 py-1 text-xs font-mono rounded transition-colors {filters.downloadedOnly
-          ? 'bg-green-500/20 text-green-400 border border-green-500/30'
-          : 'bg-white/5 text-white/60 hover:bg-white/10 border border-transparent'}"
-        onclick={() =>
-          onChange({ ...filters, downloadedOnly: !filters.downloadedOnly })}
-      >
-        <svg
-          class="w-3.5 h-3.5 inline-block"
-          viewBox="0 0 24 24"
-          fill="none"
-          stroke="currentColor"
-          stroke-width="2"
-          stroke-linecap="round"
-          stroke-linejoin="round"
+      <div class="flex flex-wrap gap-1.5">
+        <button
+          type="button"
+          class="px-2 py-1 text-xs font-mono rounded transition-colors {filters.downloadedOnly
+            ? 'bg-green-500/20 text-green-400 border border-green-500/30'
+            : 'bg-white/5 text-white/60 hover:bg-white/10 border border-transparent'}"
+          onclick={() =>
+            onChange({ ...filters, downloadedOnly: !filters.downloadedOnly })}
         >
-          <path
-            class="text-white/40"
-            d="M20 20a2 2 0 0 0 2-2V8a2 2 0 0 0-2-2h-7.9a2 2 0 0 1-1.69-.9L9.6 3.9A2 2 0 0 0 7.93 3H4a2 2 0 0 0-2 2v13a2 2 0 0 0 2 2Z"
-          />
-          <path class="text-green-400" d="m9 13 2 2 4-4" />
-        </svg>
-        <span class="ml-1">Downloaded</span>
-      </button>
+          <svg
+            class="w-3.5 h-3.5 inline-block"
+            viewBox="0 0 24 24"
+            fill="none"
+            stroke="currentColor"
+            stroke-width="2"
+            stroke-linecap="round"
+            stroke-linejoin="round"
+          >
+            <path
+              class="text-white/40"
+              d="M20 20a2 2 0 0 0 2-2V8a2 2 0 0 0-2-2h-7.9a2 2 0 0 1-1.69-.9L9.6 3.9A2 2 0 0 0 7.93 3H4a2 2 0 0 0-2 2v13a2 2 0 0 0 2 2Z"
+            />
+            <path class="text-green-400" d="m9 13 2 2 4-4" />
+          </svg>
+          <span class="ml-1">Downloaded</span>
+        </button>
+        <button
+          type="button"
+          class="px-2 py-1 text-xs font-mono rounded transition-colors {filters.readyOnly
+            ? 'bg-green-500/20 text-green-400 border border-green-500/30'
+            : 'bg-white/5 text-white/60 hover:bg-white/10 border border-transparent'}"
+          onclick={() =>
+            onChange({ ...filters, readyOnly: !filters.readyOnly })}
+        >
+          <svg
+            class="w-3.5 h-3.5 inline-block"
+            viewBox="0 0 24 24"
+            fill="none"
+            stroke="currentColor"
+            stroke-width="2"
+            stroke-linecap="round"
+            stroke-linejoin="round"
+          >
+            <circle cx="12" cy="12" r="10" />
+            <path d="m9 12 2 2 4-4" />
+          </svg>
+          <span class="ml-1">Ready</span>
+        </button>
+      </div>
     </div>
 
     <!-- Size range -->

dashboard/src/lib/components/ModelPickerGroup.svelte

@@ -32,6 +32,7 @@
     group: ModelGroup;
     isExpanded: boolean;
     isFavorite: boolean;
+    isHighlighted?: boolean;
     selectedModelId: string | null;
     canModelFit: (id: string) => boolean;
     getModelFitStatus: (id: string) => ModelFitStatus;
@@ -41,12 +42,14 @@
     onShowInfo: (group: ModelGroup) => void;
     downloadStatusMap?: Map<string, DownloadAvailability>;
     launchedAt?: number;
+    instanceStatuses?: Record<string, { status: string; statusClass: string }>;
   };
 
   let {
     group,
     isExpanded,
     isFavorite,
+    isHighlighted = false,
     selectedModelId,
     canModelFit,
     getModelFitStatus,
@@ -56,6 +59,7 @@
     onShowInfo,
     downloadStatusMap,
     launchedAt,
+    instanceStatuses = {},
   }: ModelPickerGroupProps = $props();
 
   // Group-level download status: show if any variant is downloaded
@@ -92,6 +96,12 @@
   const anyVariantFits = $derived(
     group.variants.some((v) => canModelFit(v.id)),
   );
+  // Check if any variant has an active instance (ready, loading, downloading)
+  const anyVariantHasInstance = $derived(
+    instanceStatuses
+      ? group.variants.some((v) => instanceStatuses[v.id] != null)
+      : false,
+  );
   const groupFitStatus = $derived.by((): ModelFitStatus => {
     let hasClusterCapacityOnly = false;
     for (const variant of group.variants) {
@@ -122,16 +132,36 @@
     !group.hasMultipleVariants &&
       group.variants.some((v) => v.id === selectedModelId),
   );
+
+  // Group-level instance status: show the "best" status across all variants
+  const groupInstanceStatus = $derived.by(() => {
+    if (!instanceStatuses) return null;
+    const readyStatuses = ["READY", "LOADED", "RUNNING"];
+    const loadingStatuses = ["LOADING", "WARMING UP"];
+    let bestStatus: { status: string; statusClass: string } | null = null;
+    for (const variant of group.variants) {
+      const s = instanceStatuses[variant.id];
+      if (!s) continue;
+      if (readyStatuses.includes(s.status)) return s; // Ready is best
+      if (loadingStatuses.includes(s.status) || s.status === "DOWNLOADING") {
+        bestStatus = s;
+      }
+    }
+    return bestStatus;
+  });
 </script>
 
 <div
-  class="border-b border-white/5 last:border-b-0 {!anyVariantFits
+  data-model-ids={group.variants.map((v) => v.id).join(" ")}
+  class="border-b border-white/5 last:border-b-0 {!anyVariantFits &&
+  !anyVariantHasInstance
     ? 'opacity-50'
-    : ''}"
+    : ''} {isHighlighted ? 'model-just-added' : ''}"
 >
   <!-- Main row -->
   <div
-    class="flex items-center gap-2 px-3 py-2.5 transition-colors {anyVariantFits
+    class="flex items-center gap-2 px-3 py-2.5 transition-colors {anyVariantFits ||
+    anyVariantHasInstance
       ? 'hover:bg-white/5 cursor-pointer'
       : 'cursor-not-allowed'} {isMainSelected
       ? 'bg-exo-yellow/10 border-l-2 border-exo-yellow'
@@ -141,7 +171,7 @@
         onToggleExpand();
       } else {
         const modelId = group.variants[0]?.id;
-        if (modelId && canModelFit(modelId)) {
+        if (modelId && (canModelFit(modelId) || instanceStatuses[modelId])) {
           onSelectModel(modelId);
         }
       }
@@ -155,7 +185,7 @@
           onToggleExpand();
         } else {
           const modelId = group.variants[0]?.id;
-          if (modelId && canModelFit(modelId)) {
+          if (modelId && (canModelFit(modelId) || instanceStatuses[modelId])) {
             onSelectModel(modelId);
           }
         }
@@ -346,6 +376,47 @@
       </span>
     {/if}
 
+    <!-- Instance status badge -->
+    {#if groupInstanceStatus}
+      {#if groupInstanceStatus.status === "READY" || groupInstanceStatus.status === "LOADED" || groupInstanceStatus.status === "RUNNING"}
+        <span class="flex-shrink-0" title="Running">
+          <svg
+            class="w-3 h-3 text-green-400"
+            viewBox="0 0 12 12"
+            fill="currentColor"
+          >
+            <circle cx="6" cy="6" r="5" />
+          </svg>
+        </span>
+      {:else if groupInstanceStatus.status === "DOWNLOADING"}
+        <span class="flex-shrink-0 animate-pulse" title="Downloading">
+          <svg
+            class="w-3.5 h-3.5 text-blue-400"
+            viewBox="0 0 24 24"
+            fill="none"
+            stroke="currentColor"
+            stroke-width="2"
+            stroke-linecap="round"
+            stroke-linejoin="round"
+          >
+            <path d="M21 15v4a2 2 0 01-2 2H5a2 2 0 01-2-2v-4" />
+            <polyline points="7 10 12 15 17 10" />
+            <line x1="12" y1="15" x2="12" y2="3" />
+          </svg>
+        </span>
+      {:else if groupInstanceStatus.status === "LOADING" || groupInstanceStatus.status === "WARMING UP"}
+        <span class="flex-shrink-0 animate-pulse" title="Loading">
+          <svg
+            class="w-3 h-3 text-yellow-400"
+            viewBox="0 0 12 12"
+            fill="currentColor"
+          >
+            <circle cx="6" cy="6" r="5" />
+          </svg>
+        </span>
+      {/if}
+    {/if}
+
     <!-- Check mark if selected (single-variant) -->
     {#if isMainSelected}
       <svg
@@ -420,20 +491,30 @@
       {#each group.variants as variant}
         {@const fitStatus = getModelFitStatus(variant.id)}
         {@const modelCanFit = canModelFit(variant.id)}
+        {@const variantHasInstance = instanceStatuses[variant.id] != null}
         {@const isSelected = selectedModelId === variant.id}
-        <button
-          type="button"
-          class="w-full flex items-center gap-3 px-3 py-2 pl-10 hover:bg-white/5 transition-colors text-left {!modelCanFit
+        <div
+          class="w-full flex items-center gap-3 px-3 py-2 pl-10 hover:bg-white/5 transition-colors text-left {!modelCanFit &&
+          !variantHasInstance
             ? 'opacity-50 cursor-not-allowed'
             : 'cursor-pointer'} {isSelected
             ? 'bg-exo-yellow/10 border-l-2 border-exo-yellow'
             : 'border-l-2 border-transparent'}"
-          disabled={!modelCanFit}
+          role="button"
+          tabindex="0"
           onclick={() => {
-            if (modelCanFit) {
+            if (modelCanFit || variantHasInstance) {
               onSelectModel(variant.id);
             }
           }}
+          onkeydown={(e) => {
+            if (e.key === "Enter" || e.key === " ") {
+              e.preventDefault();
+              if (modelCanFit) {
+                onSelectModel(variant.id);
+              }
+            }
+          }}
         >
           <!-- Quantization badge -->
           <span
@@ -478,6 +559,48 @@
             {/if}
           {/if}
 
+          <!-- Instance status badge -->
+          {#if instanceStatuses[variant.id]}
+            {@const instStatus = instanceStatuses[variant.id]}
+            {#if instStatus.status === "READY" || instStatus.status === "LOADED" || instStatus.status === "RUNNING"}
+              <span class="flex-shrink-0" title="Running">
+                <svg
+                  class="w-3 h-3 text-green-400"
+                  viewBox="0 0 12 12"
+                  fill="currentColor"
+                >
+                  <circle cx="6" cy="6" r="5" />
+                </svg>
+              </span>
+            {:else if instStatus.status === "DOWNLOADING"}
+              <span class="flex-shrink-0 animate-pulse" title="Downloading">
+                <svg
+                  class="w-3.5 h-3.5 text-blue-400"
+                  viewBox="0 0 24 24"
+                  fill="none"
+                  stroke="currentColor"
+                  stroke-width="2"
+                  stroke-linecap="round"
+                  stroke-linejoin="round"
+                >
+                  <path d="M21 15v4a2 2 0 01-2 2H5a2 2 0 01-2-2v-4" />
+                  <polyline points="7 10 12 15 17 10" />
+                  <line x1="12" y1="15" x2="12" y2="3" />
+                </svg>
+              </span>
+            {:else if instStatus.status === "LOADING" || instStatus.status === "WARMING UP"}
+              <span class="flex-shrink-0 animate-pulse" title="Loading">
+                <svg
+                  class="w-3 h-3 text-yellow-400"
+                  viewBox="0 0 12 12"
+                  fill="currentColor"
+                >
+                  <circle cx="6" cy="6" r="5" />
+                </svg>
+              </span>
+            {/if}
+          {/if}
+
           <!-- Check mark if selected -->
           {#if isSelected}
             <svg
@@ -490,8 +613,55 @@
               />
             </svg>
           {/if}
-        </button>
+
+          <!-- Info button -->
+          <button
+            type="button"
+            class="p-1 rounded hover:bg-white/10 transition-colors flex-shrink-0"
+            onclick={(e) => {
+              e.stopPropagation();
+              onShowInfo({
+                id: variant.id,
+                name: variant.name || variant.id,
+                capabilities: group.capabilities,
+                family: group.family,
+                variants: [variant],
+                smallestVariant: variant,
+                hasMultipleVariants: false,
+              });
+            }}
+            title="View variant details"
+          >
+            <svg
+              class="w-4 h-4 text-white/30 hover:text-white/50"
+              viewBox="0 0 24 24"
+              fill="currentColor"
+            >
+              <path
+                d="M12 2C6.48 2 2 6.48 2 12s4.48 10 10 10 10-4.48 10-10S17.52 2 12 2zm1 15h-2v-6h2v6zm0-8h-2V7h2v2z"
+              />
+            </svg>
+          </button>
+        </div>
       {/each}
     </div>
   {/if}
 </div>
+
+<style>
+  .model-just-added {
+    animation: highlightFade 4s ease-out forwards;
+  }
+
+  @keyframes highlightFade {
+    0%,
+    40% {
+      background-color: rgba(20, 83, 45, 0.25);
+      box-shadow: inset 0 0 0 1px rgba(74, 222, 128, 0.4);
+    }
+    100% {
+      background-color: transparent;
+      box-shadow: none;
+    }
+  }
+</style>

dashboard/src/lib/components/ModelPickerModal.svelte

@@ -1,4 +1,5 @@
 <script lang="ts">
+  import { tick } from "svelte";
   import { fade, fly } from "svelte/transition";
   import { cubicOut } from "svelte/easing";
   import FamilySidebar from "./FamilySidebar.svelte";
@@ -7,6 +8,7 @@
   import HuggingFaceResultItem from "./HuggingFaceResultItem.svelte";
   import { getNodesWithModelDownloaded } from "$lib/utils/downloads";
   import { getRecentEntries } from "$lib/stores/recents.svelte";
+  import { addToast } from "$lib/stores/toast.svelte";
 
   interface ModelInfo {
     id: string;
@@ -36,6 +38,7 @@
     capabilities: string[];
     sizeRange: { min: number; max: number } | null;
     downloadedOnly: boolean;
+    readyOnly: boolean;
   }
 
   interface HuggingFaceModel {
@@ -49,6 +52,11 @@
 
   type ModelFitStatus = "fits_now" | "fits_cluster_capacity" | "too_large";
 
+  export type InstanceStatus = {
+    status: string;
+    statusClass: string;
+  };
+
   type ModelPickerModalProps = {
     isOpen: boolean;
     models: ModelInfo[];
@@ -75,6 +83,7 @@
         macmon_info?: { memory?: { ram_total?: number } };
       }
     >;
+    instanceStatuses?: Record<string, InstanceStatus>;
   };
 
   let {
@@ -96,6 +105,7 @@
     usedMemoryGB,
     downloadsData,
     topologyNodes,
+    instanceStatuses = {},
   }: ModelPickerModalProps = $props();
 
   // Local state
@@ -107,6 +117,7 @@
     capabilities: [],
     sizeRange: null,
     downloadedOnly: false,
+    readyOnly: false,
   });
   let infoGroup = $state<ModelGroup | null>(null);
 
@@ -182,6 +193,13 @@
   let hfSearchDebounceTimer: ReturnType<typeof setTimeout> | null = null;
   let manualModelId = $state("");
   let addModelError = $state<string | null>(null);
+  let justAddedModelId = $state<string | null>(null);
+  let justAddedTimer: ReturnType<typeof setTimeout> | null = null;
+
+  // Inline HuggingFace search in main search bar
+  let mainSearchHfResults = $state<HuggingFaceModel[]>([]);
+  let mainSearchHfLoading = $state(false);
+  let mainSearchDebounceTimer: ReturnType<typeof setTimeout> | null = null;
 
   // Reset transient state when modal opens, but preserve tab selection
   $effect(() => {
@@ -191,6 +209,11 @@
       showFilters = false;
       manualModelId = "";
       addModelError = null;
+      justAddedModelId = null;
+      if (justAddedTimer) {
+        clearTimeout(justAddedTimer);
+        justAddedTimer = null;
+      }
     }
   });
 
@@ -205,6 +228,50 @@
     }
   });
 
+  // Inline HuggingFace search when local search returns no results
+  $effect(() => {
+    const query = searchQuery.trim();
+    const noLocalResults = filteredGroups.length === 0;
+
+    if (mainSearchDebounceTimer) {
+      clearTimeout(mainSearchDebounceTimer);
+      mainSearchDebounceTimer = null;
+    }
+
+    if (
+      selectedFamily === "huggingface" ||
+      selectedFamily === "recents" ||
+      selectedFamily === "favorites" ||
+      query.length < 2 ||
+      !noLocalResults
+    ) {
+      mainSearchHfResults = [];
+      mainSearchHfLoading = false;
+      return;
+    }
+
+    mainSearchHfLoading = true;
+    mainSearchDebounceTimer = setTimeout(async () => {
+      try {
+        const response = await fetch(
+          `/models/search?query=${encodeURIComponent(query)}&limit=10`,
+        );
+        if (response.ok) {
+          const results: HuggingFaceModel[] = await response.json();
+          mainSearchHfResults = results.filter(
+            (r) => !existingModelIds.has(r.id),
+          );
+        } else {
+          mainSearchHfResults = [];
+        }
+      } catch {
+        mainSearchHfResults = [];
+      } finally {
+        mainSearchHfLoading = false;
+      }
+    }, 500);
+  });
+
   async function fetchTrendingModels() {
     hfIsLoadingTrending = true;
     try {
@@ -265,6 +332,24 @@
     addModelError = null;
     try {
       await onAddModel(modelId);
+      // Success: show toast, switch to All Models, highlight the model
+      const shortName = modelId.split("/").pop() || modelId;
+      addToast({ type: "success", message: `Added ${shortName}` });
+      justAddedModelId = modelId;
+      selectedFamily = null;
+      searchQuery = "";
+      // Scroll to the newly added model after DOM update
+      await tick();
+      const el = document.querySelector(
+        `[data-model-ids~="${CSS.escape(modelId)}"]`,
+      );
+      el?.scrollIntoView({ behavior: "smooth", block: "center" });
+      // Clear highlight after 4 seconds
+      if (justAddedTimer) clearTimeout(justAddedTimer);
+      justAddedTimer = setTimeout(() => {
+        justAddedModelId = null;
+        justAddedTimer = null;
+      }, 4000);
     } catch (error) {
       addModelError =
         error instanceof Error ? error.message : "Failed to add model";
@@ -374,6 +459,7 @@
       "llama",
       "flux",
       "qwen-image",
+      "nemotron",
     ];
     return Array.from(families).sort((a, b) => {
       const aIdx = familyOrder.indexOf(a);
@@ -440,6 +526,16 @@
       );
     }
 
+    // Filter to ready/running models only
+    if (filters.readyOnly) {
+      result = result.filter((g) =>
+        g.variants.some((v) => {
+          const s = instanceStatuses[v.id];
+          return s && s.statusClass === "ready";
+        }),
+      );
+    }
+
     // Sort: fits-now first, then fits-cluster-capacity, then too-large
     result.sort((a, b) => {
       const getGroupFitRank = (group: ModelGroup): number => {
@@ -512,6 +608,18 @@
     );
   });
 
+  // Split filtered groups into recommended (fits_now) and others for visual separation
+  const recommendedGroups = $derived(
+    filteredGroups.filter((g) =>
+      g.variants.some((v) => getModelFitStatus(v.id) === "fits_now"),
+    ),
+  );
+  const otherGroups = $derived(
+    filteredGroups.filter(
+      (g) => !g.variants.some((v) => getModelFitStatus(v.id) === "fits_now"),
+    ),
+  );
+
   function toggleGroupExpanded(groupId: string) {
     const next = new Set(expandedGroups);
     if (next.has(groupId)) {
@@ -538,13 +646,19 @@
   }
 
   function clearFilters() {
-    filters = { capabilities: [], sizeRange: null, downloadedOnly: false };
+    filters = {
+      capabilities: [],
+      sizeRange: null,
+      downloadedOnly: false,
+      readyOnly: false,
+    };
   }
 
   const hasActiveFilters = $derived(
     filters.capabilities.length > 0 ||
       filters.sizeRange !== null ||
-      filters.downloadedOnly,
+      filters.downloadedOnly ||
+      filters.readyOnly,
   );
 </script>
 
@@ -804,6 +918,8 @@
                 {group}
                 isExpanded={expandedGroups.has(group.id)}
                 isFavorite={favorites.has(group.id)}
+                isHighlighted={justAddedModelId !== null &&
+                  group.variants.some((v) => v.id === justAddedModelId)}
                 {selectedModelId}
                 {canModelFit}
                 {getModelFitStatus}
@@ -813,6 +929,7 @@
                 onShowInfo={(g) => (infoGroup = g)}
                 downloadStatusMap={getVariantDownloadMap(group)}
                 launchedAt={recentTimestamps.get(group.variants[0]?.id ?? "")}
+                {instanceStatuses}
               />
             {/each}
           {/if}
@@ -840,11 +957,40 @@
             {/if}
           </div>
         {:else}
-          {#each filteredGroups as group}
+          <!-- Recommended for your cluster -->
+          {#if recommendedGroups.length > 0 && otherGroups.length > 0 && !searchQuery.trim()}
+            <div
+              class="sticky top-0 z-10 flex items-center gap-2 px-3 py-2 bg-green-950/60 border-b border-green-500/20 backdrop-blur-sm"
+            >
+              <svg
+                class="w-3.5 h-3.5 text-green-400 flex-shrink-0"
+                fill="none"
+                viewBox="0 0 24 24"
+                stroke="currentColor"
+                stroke-width="2"
+              >
+                <path
+                  stroke-linecap="round"
+                  stroke-linejoin="round"
+                  d="M9 12.75L11.25 15 15 9.75M21 12a9 9 0 11-18 0 9 9 0 0118 0z"
+                />
+              </svg>
+              <span
+                class="text-xs font-mono text-green-400 tracking-wider uppercase"
+                >Recommended for your cluster</span
+              >
+              <span class="text-xs font-mono text-green-400/50"
+                >— fits in available memory</span
+              >
+            </div>
+          {/if}
+          {#each recommendedGroups as group}
             <ModelPickerGroup
               {group}
               isExpanded={expandedGroups.has(group.id)}
               isFavorite={favorites.has(group.id)}
+              isHighlighted={justAddedModelId !== null &&
+                group.variants.some((v) => v.id === justAddedModelId)}
               {selectedModelId}
               {canModelFit}
               {getModelFitStatus}
@@ -853,8 +999,87 @@
               {onToggleFavorite}
               onShowInfo={(g) => (infoGroup = g)}
               downloadStatusMap={getVariantDownloadMap(group)}
+              {instanceStatuses}
             />
           {/each}
+          <!-- Other models -->
+          {#if otherGroups.length > 0 && recommendedGroups.length > 0 && !searchQuery.trim()}
+            <div
+              class="sticky top-0 z-10 flex items-center gap-2 px-3 py-2 bg-exo-dark-gray/80 border-y border-exo-medium-gray/20 backdrop-blur-sm"
+            >
+              <span
+                class="text-xs font-mono text-white/40 tracking-wider uppercase"
+                >Other models</span
+              >
+            </div>
+          {/if}
+          {#each otherGroups as group}
+            <ModelPickerGroup
+              {group}
+              isExpanded={expandedGroups.has(group.id)}
+              isFavorite={favorites.has(group.id)}
+              isHighlighted={justAddedModelId !== null &&
+                group.variants.some((v) => v.id === justAddedModelId)}
+              {selectedModelId}
+              {canModelFit}
+              {getModelFitStatus}
+              onToggleExpand={() => toggleGroupExpanded(group.id)}
+              onSelectModel={handleSelect}
+              {onToggleFavorite}
+              onShowInfo={(g) => (infoGroup = g)}
+              downloadStatusMap={getVariantDownloadMap(group)}
+              {instanceStatuses}
+            />
+          {/each}
+          <!-- Inline HuggingFace search results (shown when no local results match) -->
+          {#if filteredGroups.length === 0 && searchQuery.trim().length >= 2 && selectedFamily !== "huggingface" && selectedFamily !== "recents" && selectedFamily !== "favorites"}
+            {#if mainSearchHfLoading}
+              <div
+                class="flex items-center gap-2 px-3 py-2 border-t border-orange-400/20 bg-orange-950/20"
+              >
+                <span
+                  class="w-4 h-4 border-2 border-orange-400 border-t-transparent rounded-full animate-spin"
+                ></span>
+                <span class="text-xs font-mono text-orange-400/60"
+                  >Searching HuggingFace...</span
+                >
+              </div>
+            {:else if mainSearchHfResults.length > 0}
+              <div
+                class="sticky top-0 z-10 flex items-center gap-2 px-3 py-2 bg-orange-950/30 border-y border-orange-400/20 backdrop-blur-sm"
+              >
+                <span
+                  class="text-xs font-mono text-orange-400 tracking-wider uppercase"
+                  >From HuggingFace</span
+                >
+              </div>
+              {#each mainSearchHfResults as model}
+                <HuggingFaceResultItem
+                  {model}
+                  isAdded={existingModelIds.has(model.id)}
+                  isAdding={addingModelId === model.id}
+                  onAdd={() => handleAddModel(model.id)}
+                  onSelect={() => handleSelectHfModel(model.id)}
+                  downloadedOnNodes={downloadsData
+                    ? getNodesWithModelDownloaded(downloadsData, model.id).map(
+                        getNodeName,
+                      )
+                    : []}
+                />
+              {/each}
+              <button
+                type="button"
+                class="w-full px-3 py-2 text-xs font-mono text-orange-400/60 hover:text-orange-400 hover:bg-orange-500/10 transition-colors text-center"
+                onclick={() => {
+                  hfSearchQuery = searchQuery;
+                  searchHuggingFace(searchQuery);
+                  selectedFamily = "huggingface";
+                }}
+              >
+                See all results on Hub
+              </button>
+            {/if}
+          {/if}
         {/if}
       </div>
     </div>
@@ -875,6 +1100,11 @@
             >Downloaded</span
           >
         {/if}
+        {#if filters.readyOnly}
+          <span class="px-1.5 py-0.5 bg-green-500/20 text-green-400 rounded"
+            >Ready</span
+          >
+        {/if}
         {#if filters.sizeRange}
           <span class="px-1.5 py-0.5 bg-exo-yellow/20 text-exo-yellow rounded">
             {filters.sizeRange.min}GB - {filters.sizeRange.max}GB

dashboard/src/lib/components/PrefillProgressBar.svelte

@@ -0,0 +1,70 @@
+<script lang="ts">
+  import type { PrefillProgress } from "$lib/stores/app.svelte";
+
+  interface Props {
+    progress: PrefillProgress;
+    class?: string;
+  }
+
+  let { progress, class: className = "" }: Props = $props();
+
+  const percentage = $derived(
+    progress.total > 0
+      ? Math.round((progress.processed / progress.total) * 100)
+      : 0,
+  );
+
+  const etaText = $derived.by(() => {
+    if (progress.processed <= 0 || progress.total <= 0) return null;
+    const elapsedMs = performance.now() - progress.startedAt;
+    if (elapsedMs < 200) return null; // need a minimum sample window
+    const tokensPerMs = progress.processed / elapsedMs;
+    const remainingTokens = progress.total - progress.processed;
+    const remainingMs = remainingTokens / tokensPerMs;
+    const remainingSec = Math.ceil(remainingMs / 1000);
+    if (remainingSec <= 0) return null;
+    if (remainingSec < 60) return `~${remainingSec}s remaining`;
+    const mins = Math.floor(remainingSec / 60);
+    const secs = remainingSec % 60;
+    return `~${mins}m ${secs}s remaining`;
+  });
+
+  function formatTokenCount(count: number | undefined): string {
+    if (count == null) return "0";
+    if (count >= 1000) {
+      return `${(count / 1000).toFixed(1)}k`;
+    }
+    return count.toString();
+  }
+</script>
+
+<div class="prefill-progress {className}">
+  <div
+    class="flex items-center justify-between text-xs text-exo-light-gray mb-1"
+  >
+    <span>Processing prompt</span>
+    <span class="font-mono">
+      {formatTokenCount(progress.processed)} / {formatTokenCount(
+        progress.total,
+      )} tokens
+    </span>
+  </div>
+  <div class="h-1.5 bg-exo-black/60 rounded-full overflow-hidden">
+    <div
+      class="h-full bg-exo-yellow rounded-full transition-all duration-150 ease-out"
+      style="width: {percentage}%"
+    ></div>
+  </div>
+  <div
+    class="flex items-center justify-between text-xs text-exo-light-gray/70 mt-0.5 font-mono"
+  >
+    <span>{etaText ?? ""}</span>
+    <span>{percentage}%</span>
+  </div>
+</div>
+
+<style>
+  .prefill-progress {
+    width: 100%;
+  }
+</style>

dashboard/src/lib/components/ToastContainer.svelte

@@ -0,0 +1,117 @@
+<script lang="ts">
+  import { toasts, dismissToast, type Toast } from "$lib/stores/toast.svelte";
+  import { fly, fade } from "svelte/transition";
+  import { flip } from "svelte/animate";
+
+  const items = $derived(toasts());
+
+  const typeStyles: Record<
+    Toast["type"],
+    { border: string; icon: string; iconColor: string }
+  > = {
+    success: {
+      border: "border-l-green-500",
+      icon: "M9 12.75L11.25 15 15 9.75M21 12a9 9 0 11-18 0 9 9 0 0118 0z",
+      iconColor: "text-green-400",
+    },
+    error: {
+      border: "border-l-red-500",
+      icon: "M12 9v3.75m9-.75a9 9 0 11-18 0 9 9 0 0118 0zm-9 3.75h.008v.008H12v-.008z",
+      iconColor: "text-red-400",
+    },
+    warning: {
+      border: "border-l-yellow-500",
+      icon: "M12 9v3.75m-9.303 3.376c-.866 1.5.217 3.374 1.948 3.374h14.71c1.73 0 2.813-1.874 1.948-3.374L13.949 3.378c-.866-1.5-3.032-1.5-3.898 0L2.697 16.126z",
+      iconColor: "text-yellow-400",
+    },
+    info: {
+      border: "border-l-blue-500",
+      icon: "M11.25 11.25l.041-.02a.75.75 0 011.063.852l-.708 2.836a.75.75 0 001.063.853l.041-.021M21 12a9 9 0 11-18 0 9 9 0 0118 0zm-9-3.75h.008v.008H12V8.25z",
+      iconColor: "text-blue-400",
+    },
+  };
+</script>
+
+{#if items.length > 0}
+  <div
+    class="fixed bottom-6 right-6 z-[9999] flex flex-col gap-2 pointer-events-none"
+    role="log"
+    aria-live="polite"
+    aria-label="Notifications"
+  >
+    {#each items as toast (toast.id)}
+      {@const style = typeStyles[toast.type]}
+      <div
+        class="pointer-events-auto max-w-sm w-80 bg-exo-dark-gray/95 backdrop-blur-sm border border-exo-medium-gray/60 border-l-[3px] {style.border} rounded shadow-lg shadow-black/40"
+        in:fly={{ x: 80, duration: 250 }}
+        out:fade={{ duration: 150 }}
+        animate:flip={{ duration: 200 }}
+        role="alert"
+      >
+        <div class="flex items-start gap-3 px-4 py-3">
+          <!-- Icon -->
+          <svg
+            class="w-5 h-5 flex-shrink-0 mt-0.5 {style.iconColor}"
+            fill="none"
+            viewBox="0 0 24 24"
+            stroke-width="1.5"
+            stroke="currentColor"
+          >
+            <path
+              stroke-linecap="round"
+              stroke-linejoin="round"
+              d={style.icon}
+            />
+          </svg>
+
+          <!-- Message -->
+          <p class="flex-1 text-sm text-white/90 font-mono leading-snug">
+            {toast.message}
+          </p>
+
+          <!-- Dismiss button -->
+          <button
+            onclick={() => dismissToast(toast.id)}
+            class="flex-shrink-0 p-0.5 text-white/40 hover:text-white/80 transition-colors cursor-pointer"
+            aria-label="Dismiss notification"
+          >
+            <svg
+              class="w-4 h-4"
+              fill="none"
+              viewBox="0 0 24 24"
+              stroke-width="2"
+              stroke="currentColor"
+            >
+              <path
+                stroke-linecap="round"
+                stroke-linejoin="round"
+                d="M6 18L18 6M6 6l12 12"
+              />
+            </svg>
+          </button>
+        </div>
+
+        <!-- Auto-dismiss progress bar -->
+        {#if toast.duration > 0}
+          <div class="h-0.5 bg-white/5 rounded-b overflow-hidden">
+            <div
+              class="h-full {style.border.replace('border-l-', 'bg-')}/60"
+              style="animation: shrink {toast.duration}ms linear forwards"
+            ></div>
+          </div>
+        {/if}
+      </div>
+    {/each}
+  </div>
+{/if}
+
+<style>
+  @keyframes shrink {
+    from {
+      width: 100%;
+    }
+    to {
+      width: 0%;
+    }
+  }
+</style>

dashboard/src/lib/components/index.ts

@@ -12,3 +12,4 @@ export { default as HuggingFaceResultItem } from "./HuggingFaceResultItem.svelte
 export { default as ModelFilterPopover } from "./ModelFilterPopover.svelte";
 export { default as ModelPickerGroup } from "./ModelPickerGroup.svelte";
 export { default as ModelPickerModal } from "./ModelPickerModal.svelte";
+export { default as ChatModelSelector } from "./ChatModelSelector.svelte";

dashboard/src/lib/stores/app.svelte.ts

@@ -168,7 +168,7 @@ export interface ModelDownloadStatus {
 export interface PlacementPreview {
   model_id: string;
   sharding: "Pipeline" | "Tensor";
-  instance_meta: "MlxRing" | "MlxIbv" | "MlxJaccl";
+  instance_meta: "MlxRing" | "MlxJaccl";
   instance: unknown | null;
   memory_delta_by_node: Record<string, number> | null;
   error: string | null;
@@ -219,7 +219,6 @@ interface RawStateResponse {
     string,
     {
       MlxRingInstance?: Instance;
-      MlxIbvInstance?: Instance;
       MlxJacclInstance?: Instance;
     }
   >;
@@ -250,6 +249,11 @@ interface RawStateResponse {
   >;
   // Thunderbolt bridge cycles (nodes with bridge enabled forming loops)
   thunderboltBridgeCycles?: string[][];
+  // Disk usage per node
+  nodeDisk?: Record<
+    string,
+    { total: { inBytes: number }; available: { inBytes: number } }
+  >;
 }
 
 export interface MessageAttachment {
@@ -273,6 +277,13 @@ export interface TokenData {
   topLogprobs: TopLogprob[];
 }
 
+export interface PrefillProgress {
+  processed: number;
+  total: number;
+  /** Timestamp (performance.now()) when prefill started. */
+  startedAt: number;
+}
+
 export interface Message {
   id: string;
   role: "user" | "assistant" | "system";
@@ -306,13 +317,14 @@ const IMAGE_PARAMS_STORAGE_KEY = "exo-image-generation-params";
 export interface ImageGenerationParams {
   // Basic params
   size:
+    | "auto"
     | "512x512"
     | "768x768"
     | "1024x1024"
     | "1024x768"
     | "768x1024"
-    | "1024x1365"
-    | "1365x1024";
+    | "1024x1536"
+    | "1536x1024";
   quality: "low" | "medium" | "high";
   outputFormat: "png" | "jpeg";
   numImages: number;
@@ -336,7 +348,7 @@ export interface EditingImage {
 }
 
 const DEFAULT_IMAGE_PARAMS: ImageGenerationParams = {
-  size: "1024x1024",
+  size: "auto",
   quality: "medium",
   outputFormat: "png",
   numImages: 1,
@@ -519,6 +531,10 @@ class AppStore {
   ttftMs = $state<number | null>(null); // Time to first token in ms
   tps = $state<number | null>(null); // Tokens per second
   totalTokens = $state<number>(0); // Total tokens in current response
+  prefillProgress = $state<PrefillProgress | null>(null);
+
+  // Abort controller for stopping generation
+  private currentAbortController: AbortController | null = null;
 
   // Topology state
   topologyData = $state<TopologyData | null>(null);
@@ -564,6 +580,8 @@ class AppStore {
   debugMode = $state(false);
   topologyOnlyMode = $state(false);
   chatSidebarVisible = $state(true); // Shown by default
+  mobileChatSidebarOpen = $state(false); // Mobile drawer state
+  mobileRightSidebarOpen = $state(false); // Mobile right drawer state
 
   // Image generation params
   imageGenerationParams = $state<ImageGenerationParams>({
@@ -573,6 +591,12 @@ class AppStore {
   // Image editing state
   editingImage = $state<EditingImage | null>(null);
 
+  /** True when the backend is reachable. */
+  isConnected = $state<boolean>(true);
+  /** Number of consecutive fetch failures. */
+  private consecutiveFailures = 0;
+  private static readonly CONNECTION_LOST_THRESHOLD = 3;
+
   private fetchInterval: ReturnType<typeof setInterval> | null = null;
   private previewsInterval: ReturnType<typeof setInterval> | null = null;
   private lastConversationPersistTs = 0;
@@ -825,6 +849,10 @@ class AppStore {
     this.thinkingEnabled = conversation.enableThinking ?? true;
     this.refreshConversationModelFromInstances();
 
+    // Sync global selection to the loaded conversation's model so reactive
+    // effects in +page.svelte can determine the correct chat launch state.
+    this.selectedChatModel = conversation.modelId || "";
+
     return true;
   }
 
@@ -895,11 +923,7 @@ class AppStore {
 
     let instanceType: string | null = null;
     if (instanceTag === "MlxRingInstance") instanceType = "MLX Ring";
-    else if (
-      instanceTag === "MlxIbvInstance" ||
-      instanceTag === "MlxJacclInstance"
-    )
-      instanceType = "MLX RDMA";
+    else if (instanceTag === "MlxJacclInstance") instanceType = "MLX RDMA";
 
     let sharding: string | null = null;
     const inst = instance as {
@@ -1223,6 +1247,30 @@ class AppStore {
     this.saveChatSidebarVisibleToStorage();
   }
 
+  getMobileChatSidebarOpen(): boolean {
+    return this.mobileChatSidebarOpen;
+  }
+
+  setMobileChatSidebarOpen(open: boolean) {
+    this.mobileChatSidebarOpen = open;
+  }
+
+  toggleMobileChatSidebar() {
+    this.mobileChatSidebarOpen = !this.mobileChatSidebarOpen;
+  }
+
+  getMobileRightSidebarOpen(): boolean {
+    return this.mobileRightSidebarOpen;
+  }
+
+  setMobileRightSidebarOpen(open: boolean) {
+    this.mobileRightSidebarOpen = open;
+  }
+
+  toggleMobileRightSidebar() {
+    this.mobileRightSidebarOpen = !this.mobileRightSidebarOpen;
+  }
+
   startPolling() {
     this.fetchState();
     this.fetchInterval = setInterval(() => this.fetchState(), 1000);
@@ -1278,7 +1326,19 @@ class AppStore {
       // Thunderbolt bridge status per node
       this.nodeThunderboltBridge = data.nodeThunderboltBridge ?? {};
       this.lastUpdate = Date.now();
+      // Connection recovered
+      if (!this.isConnected) {
+        this.isConnected = true;
+      }
+      this.consecutiveFailures = 0;
     } catch (error) {
+      this.consecutiveFailures++;
+      if (
+        this.consecutiveFailures >= AppStore.CONNECTION_LOST_THRESHOLD &&
+        this.isConnected
+      ) {
+        this.isConnected = false;
+      }
       console.error("Error fetching state:", error);
     }
   }
@@ -1517,6 +1577,7 @@ class AppStore {
     // Remove messages after user message (including the user message for image requests
     // since generateImage/editImage will re-add it)
     this.messages = this.messages.slice(0, lastUserIndex);
+    this.updateActiveConversation();
 
     switch (requestType) {
       case "image-generation":
@@ -1642,11 +1703,12 @@ class AppStore {
       if (!reader) throw new Error("No response body");
 
       let fullContent = prefixText;
+      let streamedThinking = "";
       const collectedTokens: TokenData[] = [...tokensToKeep];
 
       interface ChatCompletionChunk {
         choices?: Array<{
-          delta?: { content?: string };
+          delta?: { content?: string; reasoning_content?: string };
           logprobs?: {
             content?: Array<{
               token: string;
@@ -1667,6 +1729,7 @@ class AppStore {
         (parsed) => {
           const choice = parsed.choices?.[0];
           const delta = choice?.delta?.content;
+          const thinkingDelta = choice?.delta?.reasoning_content;
 
           // Collect logprobs data
           const logprobsContent = choice?.logprobs?.content;
@@ -1685,7 +1748,11 @@ class AppStore {
             }
           }
 
-          if (delta) {
+          if (thinkingDelta) {
+            streamedThinking += thinkingDelta;
+          }
+
+          if (delta || thinkingDelta) {
             if (firstTokenTime === null) {
               firstTokenTime = performance.now();
               this.ttftMs = firstTokenTime - requestStartTime;
@@ -1699,9 +1766,14 @@ class AppStore {
               this.tps = ((tokenCount - tokensToKeep.length) / elapsed) * 1000;
             }
 
-            fullContent += delta;
-            const { displayContent, thinkingContent } =
+            if (delta) {
+              fullContent += delta;
+            }
+            const { displayContent, thinkingContent: tagThinking } =
               this.stripThinkingTags(fullContent);
+            const combinedThinking = [streamedThinking, tagThinking]
+              .filter(Boolean)
+              .join("\n\n");
 
             if (this.activeConversationId === targetConversationId) {
               this.currentResponse = displayContent;
@@ -1713,7 +1785,7 @@ class AppStore {
               messageId,
               (m) => {
                 m.content = displayContent;
-                m.thinking = thinkingContent || undefined;
+                m.thinking = combinedThinking || undefined;
                 m.tokens = [...collectedTokens];
               },
             );
@@ -1721,15 +1793,26 @@ class AppStore {
             this.persistConversation(targetConversationId);
           }
         },
+        {
+          generation_stats: (data) => {
+            const stats = data as { generation_tps: number };
+            if (stats.generation_tps > 0) {
+              this.tps = stats.generation_tps;
+            }
+          },
+        },
       );
 
       // Final update
       if (this.conversationExists(targetConversationId)) {
-        const { displayContent, thinkingContent } =
+        const { displayContent, thinkingContent: tagThinking } =
           this.stripThinkingTags(fullContent);
+        const finalThinking = [streamedThinking, tagThinking]
+          .filter(Boolean)
+          .join("\n\n");
         this.updateConversationMessage(targetConversationId, messageId, (m) => {
           m.content = displayContent;
-          m.thinking = thinkingContent || undefined;
+          m.thinking = finalThinking || undefined;
           m.tokens = [...collectedTokens];
           if (this.ttftMs !== null) m.ttftMs = this.ttftMs;
           if (this.tps !== null) m.tps = this.tps;
@@ -1805,7 +1888,7 @@ class AppStore {
           assistantMessage.id,
           (msg) => {
             msg.content =
-              "Error: No model available. Please launch an instance first.";
+              "No model is loaded yet. Select a model from the sidebar to get started — it will download and load automatically.";
           },
         );
         this.syncActiveMessagesIfNeeded(targetConversationId);
@@ -1837,11 +1920,12 @@ class AppStore {
       }
 
       let streamedContent = "";
+      let streamedThinking = "";
       const collectedTokens: TokenData[] = [];
 
       interface ChatCompletionChunk {
         choices?: Array<{
-          delta?: { content?: string };
+          delta?: { content?: string; reasoning_content?: string };
           logprobs?: {
             content?: Array<{
               token: string;
@@ -1862,6 +1946,7 @@ class AppStore {
         (parsed) => {
           const choice = parsed.choices?.[0];
           const delta = choice?.delta?.content;
+          const thinkingDelta = choice?.delta?.reasoning_content;
 
           // Collect logprobs data
           const logprobsContent = choice?.logprobs?.content;
@@ -1880,10 +1965,19 @@ class AppStore {
             }
           }
 
-          if (delta) {
-            streamedContent += delta;
-            const { displayContent, thinkingContent } =
+          if (thinkingDelta) {
+            streamedThinking += thinkingDelta;
+          }
+
+          if (delta || thinkingDelta) {
+            if (delta) {
+              streamedContent += delta;
+            }
+            const { displayContent, thinkingContent: tagThinking } =
               this.stripThinkingTags(streamedContent);
+            const combinedThinking = [streamedThinking, tagThinking]
+              .filter(Boolean)
+              .join("\n\n");
 
             // Only update currentResponse if target conversation is active
             if (this.activeConversationId === targetConversationId) {
@@ -1896,7 +1990,7 @@ class AppStore {
               assistantMessage.id,
               (msg) => {
                 msg.content = displayContent;
-                msg.thinking = thinkingContent || undefined;
+                msg.thinking = combinedThinking || undefined;
                 msg.tokens = [...collectedTokens];
               },
             );
@@ -1904,18 +1998,29 @@ class AppStore {
             this.persistConversation(targetConversationId);
           }
         },
+        {
+          generation_stats: (data) => {
+            const stats = data as { generation_tps: number };
+            if (stats.generation_tps > 0) {
+              this.tps = stats.generation_tps;
+            }
+          },
+        },
       );
 
       // Final cleanup of the message (if conversation still exists)
       if (this.conversationExists(targetConversationId)) {
-        const { displayContent, thinkingContent } =
+        const { displayContent, thinkingContent: tagThinking } =
           this.stripThinkingTags(streamedContent);
+        const finalThinking = [streamedThinking, tagThinking]
+          .filter(Boolean)
+          .join("\n\n");
         this.updateConversationMessage(
           targetConversationId,
           assistantMessage.id,
           (msg) => {
             msg.content = displayContent;
-            msg.thinking = thinkingContent || undefined;
+            msg.thinking = finalThinking || undefined;
             msg.tokens = [...collectedTokens];
           },
         );
@@ -2004,6 +2109,7 @@ class AppStore {
     reader: ReadableStreamDefaultReader<Uint8Array>,
     targetConversationId: string,
     onChunk: (parsed: T) => void,
+    onEvent?: Record<string, (data: unknown) => void>,
   ): Promise<void> {
     const decoder = new TextDecoder();
     let buffer = "";
@@ -2024,6 +2130,24 @@ class AppStore {
         const trimmed = line.trim();
         if (!trimmed) continue;
 
+        // Handle SSE comments (": key json") for prefill progress etc.
+        if (trimmed.startsWith(": ") && onEvent) {
+          const comment = trimmed.slice(2);
+          const spaceIdx = comment.indexOf(" ");
+          if (spaceIdx > 0) {
+            const key = comment.slice(0, spaceIdx);
+            if (onEvent[key]) {
+              try {
+                const parsed = JSON.parse(comment.slice(spaceIdx + 1));
+                onEvent[key](parsed);
+              } catch {
+                // Skip malformed JSON in comment
+              }
+            }
+          }
+          continue;
+        }
+
         if (trimmed.startsWith("data: ")) {
           const data = trimmed.slice(6);
           if (data === "[DONE]") continue;
@@ -2243,7 +2367,7 @@ class AppStore {
       const modelToUse = this.getModelForRequest();
       if (!modelToUse) {
         throw new Error(
-          "No model selected and no running instances available. Please launch an instance first.",
+          "No model is loaded yet. Select a model from the sidebar to get started — it will download and load automatically.",
         );
       }
 
@@ -2255,6 +2379,9 @@ class AppStore {
       let firstTokenTime: number | null = null;
       let tokenCount = 0;
 
+      const abortController = new AbortController();
+      this.currentAbortController = abortController;
+
       const response = await fetch("/v1/chat/completions", {
         method: "POST",
         headers: {
@@ -2271,6 +2398,7 @@ class AppStore {
             enable_thinking: enableThinking,
           }),
         }),
+        signal: abortController.signal,
       });
 
       if (!response.ok) {
@@ -2284,10 +2412,11 @@ class AppStore {
       }
 
       let streamedContent = "";
-
+      let streamedThinking = "";
+      let serverTpsReceived = false;
       interface ChatCompletionChunk {
         choices?: Array<{
-          delta?: { content?: string };
+          delta?: { content?: string; reasoning_content?: string };
           logprobs?: {
             content?: Array<{
               token: string;
@@ -2308,8 +2437,14 @@ class AppStore {
         reader,
         targetConversationId,
         (parsed) => {
+          // Clear prefill progress when first token data arrives
+          if (this.prefillProgress) {
+            this.prefillProgress = null;
+          }
+
           const choice = parsed.choices?.[0];
           const tokenContent = choice?.delta?.content;
+          const thinkingContent = choice?.delta?.reasoning_content;
 
           // Collect logprobs data
           const logprobsContent = choice?.logprobs?.content;
@@ -2328,7 +2463,11 @@ class AppStore {
             }
           }
 
-          if (tokenContent) {
+          if (thinkingContent) {
+            streamedThinking += thinkingContent;
+          }
+
+          if (tokenContent || thinkingContent) {
             // Track first token for TTFT
             if (firstTokenTime === null) {
               firstTokenTime = performance.now();
@@ -2339,17 +2478,21 @@ class AppStore {
             tokenCount += 1;
             this.totalTokens = tokenCount;
 
-            // Update real-time TPS during streaming
             if (firstTokenTime !== null && tokenCount > 1) {
               const elapsed = performance.now() - firstTokenTime;
               this.tps = (tokenCount / elapsed) * 1000;
             }
 
-            streamedContent += tokenContent;
+            if (tokenContent) {
+              streamedContent += tokenContent;
+            }
 
-            // Strip thinking tags for display and extract thinking content
-            const { displayContent, thinkingContent } =
+            // Use stripThinkingTags as fallback for any <think> tags still in content
+            const { displayContent, thinkingContent: tagThinking } =
               this.stripThinkingTags(streamedContent);
+            const combinedThinking = [streamedThinking, tagThinking]
+              .filter(Boolean)
+              .join("\n\n");
 
             // Only update currentResponse if target conversation is active
             if (this.activeConversationId === targetConversationId) {
@@ -2362,7 +2505,7 @@ class AppStore {
               assistantMessage.id,
               (msg) => {
                 msg.content = displayContent;
-                msg.thinking = thinkingContent || undefined;
+                msg.thinking = combinedThinking || undefined;
                 msg.tokens = [...collectedTokens];
               },
             );
@@ -2370,24 +2513,54 @@ class AppStore {
             this.persistConversation(targetConversationId);
           }
         },
+        {
+          prefill_progress: (data) => {
+            // TaggedModel wraps as {"PrefillProgressChunk": {...}}
+            // model_dump_json() uses snake_case (by_alias defaults to False)
+            const raw = data as Record<string, unknown>;
+            const inner = (raw["PrefillProgressChunk"] ?? raw) as {
+              processed_tokens: number;
+              total_tokens: number;
+            };
+            this.prefillProgress = {
+              processed: inner.processed_tokens,
+              total: inner.total_tokens,
+              startedAt: this.prefillProgress?.startedAt ?? performance.now(),
+            };
+          },
+          generation_stats: (data) => {
+            const stats = data as { generation_tps: number };
+
+            if (stats.generation_tps > 0) {
+              this.tps = stats.generation_tps;
+              serverTpsReceived = true;
+            }
+          },
+        },
       );
 
-      // Calculate final TPS
-      if (firstTokenTime !== null && tokenCount > 1) {
+      // Clear prefill progress after stream ends
+      this.prefillProgress = null;
+
+      // Use server-side TPS if available, otherwise fall back to client-side
+      if (!serverTpsReceived && firstTokenTime !== null && tokenCount > 1) {
         const totalGenerationTime = performance.now() - firstTokenTime;
-        this.tps = (tokenCount / totalGenerationTime) * 1000; // tokens per second
+        this.tps = (tokenCount / totalGenerationTime) * 1000;
       }
 
       // Final cleanup of the message (if conversation still exists)
       if (this.conversationExists(targetConversationId)) {
-        const { displayContent, thinkingContent } =
+        const { displayContent, thinkingContent: tagThinking } =
           this.stripThinkingTags(streamedContent);
+        const finalThinking = [streamedThinking, tagThinking]
+          .filter(Boolean)
+          .join("\n\n");
         this.updateConversationMessage(
           targetConversationId,
           assistantMessage.id,
           (msg) => {
             msg.content = displayContent;
-            msg.thinking = thinkingContent || undefined;
+            msg.thinking = finalThinking || undefined;
             msg.tokens = [...collectedTokens];
             // Store performance metrics on the message
             if (this.ttftMs !== null) {
@@ -2402,20 +2575,31 @@ class AppStore {
         this.persistConversation(targetConversationId);
       }
     } catch (error) {
-      console.error("Error sending message:", error);
-      this.handleStreamingError(
-        error,
-        targetConversationId,
-        assistantMessage.id,
-        "Failed to get response",
-      );
+      if (error instanceof DOMException && error.name === "AbortError") {
+        // User stopped generation — not an error
+      } else {
+        console.error("Error sending message:", error);
+        this.handleStreamingError(
+          error,
+          targetConversationId,
+          assistantMessage.id,
+          "Failed to get response",
+        );
+      }
     } finally {
+      this.currentAbortController = null;
+      this.prefillProgress = null;
       this.isLoading = false;
       this.currentResponse = "";
       this.saveConversationsToStorage();
     }
   }
 
+  stopGeneration(): void {
+    this.currentAbortController?.abort();
+    this.currentAbortController = null;
+  }
+
   /**
    * Generate an image using the image generation API
    */
@@ -2466,6 +2650,9 @@ class AppStore {
     this.syncActiveMessagesIfNeeded(targetConversationId);
     this.saveConversationsToStorage();
 
+    const abortController = new AbortController();
+    this.currentAbortController = abortController;
+
     try {
       // Determine the model to use
       const model = this.getModelForRequest(modelId);
@@ -2520,6 +2707,7 @@ class AppStore {
           "Content-Type": "application/json",
         },
         body: JSON.stringify(requestBody),
+        signal: abortController.signal,
       });
 
       if (!response.ok) {
@@ -2659,14 +2847,27 @@ class AppStore {
         );
       }
     } catch (error) {
-      console.error("Error generating image:", error);
-      this.handleStreamingError(
-        error,
-        targetConversationId,
-        assistantMessage.id,
-        "Failed to generate image",
-      );
+      if (abortController.signal.aborted) {
+        this.updateConversationMessage(
+          targetConversationId,
+          assistantMessage.id,
+          (msg) => {
+            msg.content = "Cancelled";
+            msg.attachments = [];
+          },
+        );
+        this.syncActiveMessagesIfNeeded(targetConversationId);
+      } else {
+        console.error("Error generating image:", error);
+        this.handleStreamingError(
+          error,
+          targetConversationId,
+          assistantMessage.id,
+          "Failed to generate image",
+        );
+      }
     } finally {
+      this.currentAbortController = null;
       this.isLoading = false;
       this.saveConversationsToStorage();
     }
@@ -2730,6 +2931,9 @@ class AppStore {
     // Clear editing state
     this.editingImage = null;
 
+    const abortController = new AbortController();
+    this.currentAbortController = abortController;
+
     try {
       // Determine the model to use
       const model = this.getModelForRequest(modelId);
@@ -2791,6 +2995,7 @@ class AppStore {
       const apiResponse = await fetch("/v1/images/edits", {
         method: "POST",
         body: formData,
+        signal: abortController.signal,
       });
 
       if (!apiResponse.ok) {
@@ -2891,14 +3096,27 @@ class AppStore {
         );
       }
     } catch (error) {
-      console.error("Error editing image:", error);
-      this.handleStreamingError(
-        error,
-        targetConversationId,
-        assistantMessage.id,
-        "Failed to edit image",
-      );
+      if (abortController.signal.aborted) {
+        this.updateConversationMessage(
+          targetConversationId,
+          assistantMessage.id,
+          (msg) => {
+            msg.content = "Cancelled";
+            msg.attachments = [];
+          },
+        );
+        this.syncActiveMessagesIfNeeded(targetConversationId);
+      } else {
+        console.error("Error editing image:", error);
+        this.handleStreamingError(
+          error,
+          targetConversationId,
+          assistantMessage.id,
+          "Failed to edit image",
+        );
+      }
     } finally {
+      this.currentAbortController = null;
       this.isLoading = false;
       this.saveConversationsToStorage();
     }
@@ -3024,6 +3242,23 @@ class AppStore {
     return (await response.json()) as TraceStatsResponse;
   }
 
+  /**
+   * Delete traces by task IDs
+   */
+  async deleteTraces(
+    taskIds: string[],
+  ): Promise<{ deleted: string[]; notFound: string[] }> {
+    const response = await fetch("/v1/traces/delete", {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({ taskIds }),
+    });
+    if (!response.ok) {
+      throw new Error(`Failed to delete traces: ${response.status}`);
+    }
+    return await response.json();
+  }
+
   /**
    * Get the URL for the raw trace file (for Perfetto)
    */
@@ -3042,6 +3277,7 @@ export const isLoading = () => appStore.isLoading;
 export const ttftMs = () => appStore.ttftMs;
 export const tps = () => appStore.tps;
 export const totalTokens = () => appStore.totalTokens;
+export const prefillProgress = () => appStore.prefillProgress;
 export const topologyData = () => appStore.topologyData;
 export const instances = () => appStore.instances;
 export const runners = () => appStore.runners;
@@ -3059,6 +3295,7 @@ export const topologyOnlyMode = () => appStore.getTopologyOnlyMode();
 export const chatSidebarVisible = () => appStore.getChatSidebarVisible();
 
 // Actions
+export const stopGeneration = () => appStore.stopGeneration();
 export const startChat = () => appStore.startChat();
 export const sendMessage = (
   content: string,
@@ -3129,8 +3366,23 @@ export const toggleChatSidebarVisible = () =>
   appStore.toggleChatSidebarVisible();
 export const setChatSidebarVisible = (visible: boolean) =>
   appStore.setChatSidebarVisible(visible);
+
+// Mobile sidebar state
+export const mobileChatSidebarOpen = () => appStore.mobileChatSidebarOpen;
+export const toggleMobileChatSidebar = () => appStore.toggleMobileChatSidebar();
+export const setMobileChatSidebarOpen = (open: boolean) =>
+  appStore.setMobileChatSidebarOpen(open);
+export const mobileRightSidebarOpen = () => appStore.mobileRightSidebarOpen;
+export const toggleMobileRightSidebar = () =>
+  appStore.toggleMobileRightSidebar();
+export const setMobileRightSidebarOpen = (open: boolean) =>
+  appStore.setMobileRightSidebarOpen(open);
+
 export const refreshState = () => appStore.fetchState();
 
+// Connection status
+export const isConnected = () => appStore.isConnected;
+
 // Node identities (for OS version mismatch detection)
 export const nodeIdentities = () => appStore.nodeIdentities;
 
@@ -3162,3 +3414,5 @@ export const fetchTraceStats = (taskId: string) =>
   appStore.fetchTraceStats(taskId);
 export const getTraceRawUrl = (taskId: string) =>
   appStore.getTraceRawUrl(taskId);
+export const deleteTraces = (taskIds: string[]) =>
+  appStore.deleteTraces(taskIds);

dashboard/src/lib/stores/toast.svelte.ts

@@ -0,0 +1,87 @@
+/**
+ * Toast notification store - Global notification system for the EXO dashboard.
+ *
+ * Usage:
+ *   import { addToast, dismissToast, toasts } from "$lib/stores/toast.svelte";
+ *   addToast({ type: "success", message: "Model launched" });
+ *   addToast({ type: "error", message: "Connection lost", persistent: true });
+ */
+
+type ToastType = "success" | "error" | "warning" | "info";
+
+export interface Toast {
+  id: string;
+  type: ToastType;
+  message: string;
+  /** Auto-dismiss after this many ms. 0 = persistent (must be dismissed manually). */
+  duration: number;
+  createdAt: number;
+}
+
+interface ToastInput {
+  type: ToastType;
+  message: string;
+  /** If true, toast stays until manually dismissed. Default: false. */
+  persistent?: boolean;
+  /** Auto-dismiss duration in ms. Default: 4000 for success/info, 6000 for error/warning. */
+  duration?: number;
+}
+
+const DEFAULT_DURATIONS: Record<ToastType, number> = {
+  success: 4000,
+  info: 4000,
+  warning: 6000,
+  error: 6000,
+};
+
+let toastList = $state<Toast[]>([]);
+const timers = new Map<string, ReturnType<typeof setTimeout>>();
+
+function generateId(): string {
+  return `toast-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`;
+}
+
+export function addToast(input: ToastInput): string {
+  const id = generateId();
+  const duration = input.persistent
+    ? 0
+    : (input.duration ?? DEFAULT_DURATIONS[input.type]);
+
+  const toast: Toast = {
+    id,
+    type: input.type,
+    message: input.message,
+    duration,
+    createdAt: Date.now(),
+  };
+
+  toastList = [...toastList, toast];
+
+  if (duration > 0) {
+    const timer = setTimeout(() => dismissToast(id), duration);
+    timers.set(id, timer);
+  }
+
+  return id;
+}
+
+export function dismissToast(id: string): void {
+  const timer = timers.get(id);
+  if (timer) {
+    clearTimeout(timer);
+    timers.delete(id);
+  }
+  toastList = toastList.filter((t) => t.id !== id);
+}
+
+/** Dismiss all toasts matching a message (useful for dedup). */
+export function dismissByMessage(message: string): void {
+  const matching = toastList.filter((t) => t.message === message);
+  for (const t of matching) {
+    dismissToast(t.id);
+  }
+}
+
+export function toasts(): Toast[] {
+  return toastList;
+}

dashboard/src/routes/+layout.svelte

@@ -1,5 +1,7 @@
 <script lang="ts">
   import "../app.css";
+  import ToastContainer from "$lib/components/ToastContainer.svelte";
+  import ConnectionBanner from "$lib/components/ConnectionBanner.svelte";
 
   let { children } = $props();
 </script>
@@ -10,5 +12,7 @@
 </svelte:head>
 
 <div class="min-h-screen bg-background text-foreground">
+  <ConnectionBanner />
   {@render children?.()}
+  <ToastContainer />
 </div>

dashboard/src/routes/traces/+page.svelte

@@ -3,6 +3,7 @@
   import {
     listTraces,
     getTraceRawUrl,
+    deleteTraces,
     type TraceListItem,
   } from "$lib/stores/app.svelte";
   import HeaderNav from "$lib/components/HeaderNav.svelte";
@@ -10,6 +11,51 @@
   let traces = $state<TraceListItem[]>([]);
   let loading = $state(true);
   let error = $state<string | null>(null);
+  let selectedIds = $state<Set<string>>(new Set());
+  let deleting = $state(false);
+
+  let allSelected = $derived(
+    traces.length > 0 && selectedIds.size === traces.length,
+  );
+
+  function toggleSelect(taskId: string) {
+    const next = new Set(selectedIds);
+    if (next.has(taskId)) {
+      next.delete(taskId);
+    } else {
+      next.add(taskId);
+    }
+    selectedIds = next;
+  }
+
+  function toggleSelectAll() {
+    if (allSelected) {
+      selectedIds = new Set();
+    } else {
+      selectedIds = new Set(traces.map((t) => t.taskId));
+    }
+  }
+
+  async function handleDelete() {
+    if (selectedIds.size === 0) return;
+    const count = selectedIds.size;
+    if (
+      !confirm(
+        `Delete ${count} trace${count === 1 ? "" : "s"}? This cannot be undone.`,
+      )
+    )
+      return;
+    deleting = true;
+    try {
+      await deleteTraces([...selectedIds]);
+      selectedIds = new Set();
+      await refresh();
+    } catch (e) {
+      error = e instanceof Error ? e.message : "Failed to delete traces";
+    } finally {
+      deleting = false;
+    }
+  }
 
   function formatBytes(bytes: number): string {
     if (!bytes || bytes <= 0) return "0B";
@@ -109,6 +155,16 @@
         </h1>
       </div>
       <div class="flex items-center gap-3">
+        {#if selectedIds.size > 0}
+          <button
+            type="button"
+            class="text-xs font-mono text-red-400 hover:text-red-300 transition-colors uppercase border border-red-500/40 px-2 py-1 rounded"
+            onclick={handleDelete}
+            disabled={deleting}
+          >
+            {deleting ? "Deleting..." : `Delete (${selectedIds.size})`}
+          </button>
+        {/if}
         <button
           type="button"
           class="text-xs font-mono text-exo-light-gray hover:text-exo-yellow transition-colors uppercase border border-exo-medium-gray/40 px-2 py-1 rounded"
@@ -143,14 +199,41 @@
       </div>
     {:else}
       <div class="space-y-3">
+        <div class="flex items-center gap-2 px-1">
+          <button
+            type="button"
+            class="text-xs font-mono uppercase transition-colors {allSelected
+              ? 'text-exo-yellow'
+              : 'text-exo-light-gray hover:text-exo-yellow'}"
+            onclick={toggleSelectAll}
+          >
+            {allSelected ? "Deselect all" : "Select all"}
+          </button>
+        </div>
         {#each traces as trace}
+          {@const isSelected = selectedIds.has(trace.taskId)}
+          <!-- svelte-ignore a11y_no_static_element_interactions -->
           <div
-            class="rounded border border-exo-medium-gray/30 bg-exo-black/30 p-4 flex items-center justify-between gap-4"
+            role="button"
+            tabindex="0"
+            class="w-full text-left rounded border-l-2 border-r border-t border-b transition-all p-4 flex items-center justify-between gap-4 cursor-pointer {isSelected
+              ? 'bg-exo-yellow/10 border-l-exo-yellow border-r-exo-medium-gray/30 border-t-exo-medium-gray/30 border-b-exo-medium-gray/30'
+              : 'bg-exo-black/30 border-l-transparent border-r-exo-medium-gray/30 border-t-exo-medium-gray/30 border-b-exo-medium-gray/30 hover:bg-white/[0.03]'}"
+            onclick={() => toggleSelect(trace.taskId)}
+            onkeydown={(e) => {
+              if (e.key === "Enter" || e.key === " ") {
+                e.preventDefault();
+                toggleSelect(trace.taskId);
+              }
+            }}
           >
             <div class="min-w-0 flex-1">
               <a
                 href="#/traces/{trace.taskId}"
-                class="text-sm font-mono text-white hover:text-exo-yellow transition-colors truncate block"
+                class="text-sm font-mono transition-colors truncate block {isSelected
+                  ? 'text-exo-yellow'
+                  : 'text-white hover:text-exo-yellow'}"
+                onclick={(e) => e.stopPropagation()}
               >
                 {trace.taskId}
               </a>
@@ -160,7 +243,11 @@
                 )}
               </div>
             </div>
-            <div class="flex items-center gap-2 shrink-0">
+            <!-- svelte-ignore a11y_click_events_have_key_events -->
+            <div
+              class="flex items-center gap-2 shrink-0"
+              onclick={(e) => e.stopPropagation()}
+            >
               <a
                 href="#/traces/{trace.taskId}"
                 class="text-xs font-mono text-exo-light-gray hover:text-exo-yellow transition-colors uppercase border border-exo-medium-gray/40 px-2 py-1 rounded"

docs/api.md

@@ -4,7 +4,7 @@ This document describes the REST API exposed by the **EXO** service, as implemen
 
 `src/exo/master/api.py`
 
-The API is used to manage model instances in the cluster, inspect cluster state, and perform inference using an OpenAI-compatible interface.
+The API is used to manage model instances in the cluster, inspect cluster state, and perform inference using multiple API-compatible interfaces.
 
 Base URL example:
 
@@ -144,8 +144,59 @@ Placement result.
 
 Returns the list of available models and their metadata.
 
+**Query parameters:**
+
+* `status`: string (optional) - Filter by `downloaded` to show only downloaded models
+
 **Response:**
-Array of model descriptors.
+Array of model descriptors including `is_custom` field for custom HuggingFace models.
+
+### Add Custom Model
+
+**POST** `/models/add`
+
+Add a custom model from HuggingFace hub.
+
+**Request body (example):**
+
+```json
+{
+  "model_id": "mlx-community/my-custom-model"
+}
+```
+
+**Response:**
+Model descriptor for the added model.
+
+**Security note:**
+Models with `trust_remote_code` enabled in their configuration require explicit opt-in (default is false) for security.
+
+### Delete Custom Model
+
+**DELETE** `/models/custom/{model_id}`
+
+Delete a user-added custom model card.
+
+**Path parameters:**
+
+* `model_id`: string, ID of the custom model to delete
+
+**Response:**
+Confirmation JSON with deleted model ID.
+
+### Search Models
+
+**GET** `/models/search`
+
+Search HuggingFace Hub for mlx-community models.
+
+**Query parameters:**
+
+* `query`: string (optional) - Search query
+* `limit`: integer (default: 20) - Maximum number of results
+
+**Response:**
+Array of HuggingFace model search results.
 
 ## 4. Inference / Chat Completions
 
@@ -168,9 +219,123 @@ Executes a chat completion request using an OpenAI-compatible schema. Supports s
 }
 ```
 
+**Request parameters:**
+
+* `model`: string, required - Model ID to use
+* `messages`: array, required - Conversation messages
+* `stream`: boolean (default: false) - Enable streaming responses
+* `max_tokens`: integer (optional) - Maximum tokens to generate
+* `temperature`: float (optional) - Sampling temperature
+* `top_p`: float (optional) - Nucleus sampling parameter
+* `top_k`: integer (optional) - Top-k sampling parameter
+* `stop`: string or array (optional) - Stop sequences
+* `seed`: integer (optional) - Random seed for reproducibility
+* `enable_thinking`: boolean (optional) - Enable thinking mode for capable models (DeepSeek V3.1, Qwen3, GLM-4.7)
+* `tools`: array (optional) - Tool definitions for function calling
+* `logprobs`: boolean (optional) - Return log probabilities
+* `top_logprobs`: integer (optional) - Number of top log probabilities to return
+
 **Response:**
 OpenAI-compatible chat completion response.
 
+**Streaming response format:**
+When `stream=true`, returns Server-Sent Events (SSE) with format:
+
+```
+data: {"id":"...","object":"chat.completion","created":...,"model":"...","choices":[...]}
+
+data: [DONE]
+```
+
+**Non-streaming response includes usage statistics:**
+
+```json
+{
+  "id": "...",
+  "object": "chat.completion",
+  "created": 1234567890,
+  "model": "llama-3.2-1b",
+  "choices": [{
+    "index": 0,
+    "message": {
+      "role": "assistant",
+      "content": "Hello! How can I help you?"
+    },
+    "finish_reason": "stop"
+  }],
+  "usage": {
+    "prompt_tokens": 15,
+    "completion_tokens": 8,
+    "total_tokens": 23
+  }
+}
+```
+
+**Cancellation:**
+You can cancel an active generation by closing the HTTP connection. The server detects the disconnection and stops processing.
+
+### Claude Messages API
+
+**POST** `/v1/messages`
+
+Executes a chat completion request using the Claude Messages API format. Supports streaming and non-streaming modes.
+
+**Request body (example):**
+
+```json
+{
+  "model": "llama-3.2-1b",
+  "messages": [
+    { "role": "user", "content": "Hello" }
+  ],
+  "max_tokens": 1024,
+  "stream": false
+}
+```
+
+**Streaming response format:**
+When `stream=true`, returns Server-Sent Events with Claude-specific event types:
+
+* `message_start` - Message generation started
+* `content_block_start` - Content block started
+* `content_block_delta` - Incremental content chunk
+* `content_block_stop` - Content block completed
+* `message_delta` - Message metadata updates
+* `message_stop` - Message generation completed
+
+**Response:**
+Claude-compatible messages response.
+
+### OpenAI Responses API
+
+**POST** `/v1/responses`
+
+Executes a chat completion request using the OpenAI Responses API format. Supports streaming and non-streaming modes.
+
+**Request body (example):**
+
+```json
+{
+  "model": "llama-3.2-1b",
+  "messages": [
+    { "role": "user", "content": "Hello" }
+  ],
+  "stream": false
+}
+```
+
+**Streaming response format:**
+When `stream=true`, returns Server-Sent Events with response-specific event types:
+
+* `response.created` - Response generation started
+* `response.in_progress` - Response is being generated
+* `response.output_item.added` - New output item added
+* `response.output_item.done` - Output item completed
+* `response.done` - Response generation completed
+
+**Response:**
+OpenAI Responses API-compatible response.
+
 ### Benchmarked Chat Completions
 
 **POST** `/bench/chat/completions`
@@ -181,26 +346,177 @@ Same as `/v1/chat/completions`, but also returns performance and generation stat
 Same schema as `/v1/chat/completions`.
 
 **Response:**
-Chat completion plus benchmarking metrics.
+Chat completion plus benchmarking metrics including:
 
-## 5. Image Generation & Editing
+* `prompt_tps` - Tokens per second during prompt processing
+* `generation_tps` - Tokens per second during generation
+* `prompt_tokens` - Number of prompt tokens
+* `generation_tokens` - Number of generated tokens
+* `peak_memory_usage` - Peak memory used during generation
+
+### Cancel Command
+
+**POST** `/v1/cancel/{command_id}`
+
+Cancels an active generation command (text or image). Notifies workers and closes the stream.
+
+**Path parameters:**
+
+* `command_id`: string, ID of the command to cancel
+
+**Response (example):**
+
+```json
+{
+  "message": "Command cancelled.",
+  "command_id": "cmd-abc-123"
+}
+```
+
+Returns 404 if the command is not found or already completed.
+
+## 5. Ollama API Compatibility
+
+EXO provides Ollama API compatibility for tools like OpenWebUI.
+
+### Ollama Chat
+
+**POST** `/ollama/api/chat`
+**POST** `/ollama/api/api/chat` (alias)
+**POST** `/ollama/api/v1/chat` (alias)
+
+Execute a chat request using Ollama API format.
+
+**Request body (example):**
+
+```json
+{
+  "model": "llama-3.2-1b",
+  "messages": [
+    { "role": "user", "content": "Hello" }
+  ],
+  "stream": false
+}
+```
+
+**Response:**
+Ollama-compatible chat response.
+
+### Ollama Generate
+
+**POST** `/ollama/api/generate`
+
+Execute a text generation request using Ollama API format.
+
+**Request body (example):**
+
+```json
+{
+  "model": "llama-3.2-1b",
+  "prompt": "Hello",
+  "stream": false
+}
+```
+
+**Response:**
+Ollama-compatible generation response.
+
+### Ollama Tags
+
+**GET** `/ollama/api/tags`
+**GET** `/ollama/api/api/tags` (alias)
+**GET** `/ollama/api/v1/tags` (alias)
+
+Returns list of downloaded models in Ollama tags format.
+
+**Response:**
+Array of model tags with metadata.
+
+### Ollama Show
+
+**POST** `/ollama/api/show`
+
+Returns model information in Ollama show format.
+
+**Request body:**
+
+```json
+{
+  "name": "llama-3.2-1b"
+}
+```
+
+**Response:**
+Model details including modelfile and family.
+
+### Ollama PS
+
+**GET** `/ollama/api/ps`
+
+Returns list of running models (active instances).
+
+**Response:**
+Array of active model instances.
+
+### Ollama Version
+
+**GET** `/ollama/api/version`
+**HEAD** `/ollama/` (alias)
+**HEAD** `/ollama/api/version` (alias)
+
+Returns version information for Ollama API compatibility.
+
+**Response:**
+
+```json
+{
+  "version": "exo v1.0"
+}
+```
+
+## 6. Image Generation & Editing
 
 ### Image Generation
 
 **POST** `/v1/images/generations`
 
-Executes an image generation request using an OpenAI-compatible schema with additional advanced_params.
+Executes an image generation request using an OpenAI-compatible schema with additional advanced_params. Supports both streaming and non-streaming modes.
 
 **Request body (example):**
 
 ```json
 {
   "prompt": "a robot playing chess",
-  "model": "flux-dev",
+  "model": "exolabs/FLUX.1-dev",
+  "n": 1,
+  "size": "1024x1024",
   "stream": false,
+  "response_format": "b64_json"
 }
 ```
 
+**Request parameters:**
+
+* `prompt`: string, required - Text description of the image
+* `model`: string, required - Image model ID
+* `n`: integer (default: 1) - Number of images to generate
+* `size`: string (default: "auto") - Image dimensions. Supported sizes:
+  - `512x512`
+  - `768x768`
+  - `1024x768`
+  - `768x1024`
+  - `1024x1024`
+  - `1024x1536`
+  - `1536x1024`
+  - `1024x1365`
+  - `1365x1024`
+* `stream`: boolean (default: false) - Enable streaming for partial images
+* `partial_images`: integer (default: 0) - Number of partial images to stream during generation
+* `response_format`: string (default: "b64_json") - Either `url` or `b64_json`
+* `quality`: string (default: "medium") - Either `high`, `medium`, or `low`
+* `output_format`: string (default: "png") - Either `png`, `jpeg`, or `webp`
+* `advanced_params`: object (optional) - Advanced generation parameters
+
 **Advanced Parameters (`advanced_params`):**
 
 | Parameter | Type | Constraints | Description |
@@ -210,8 +526,54 @@ Executes an image generation request using an OpenAI-compatible schema with addi
 | `guidance` | float | 1.0-20.0 | Classifier-free guidance scale |
 | `negative_prompt` | string | - | Text describing what to avoid in the image |
 
+**Non-streaming response:**
+
+```json
+{
+  "created": 1234567890,
+  "data": [
+    {
+      "b64_json": "iVBORw0KGgoAAAANSUhEUgAA...",
+      "url": null
+    }
+  ]
+}
+```
+
+**Streaming response format:**
+When `stream=true` and `partial_images > 0`, returns Server-Sent Events:
+
+```
+data: {"type":"partial","image_index":0,"partial_index":1,"total_partials":5,"format":"png","data":{"b64_json":"..."}}
+
+data: {"type":"final","image_index":0,"format":"png","data":{"b64_json":"..."}}
+
+data: [DONE]
+```
+
+### Image Editing
+
+**POST** `/v1/images/edits`
+
+Executes an image editing request (img2img) using FLUX.1-Kontext-dev or similar models.
+
+**Request (multipart/form-data):**
+
+* `image`: file, required - Input image to edit
+* `prompt`: string, required - Text description of desired changes
+* `model`: string, required - Image editing model ID (e.g., `exolabs/FLUX.1-Kontext-dev`)
+* `n`: integer (default: 1) - Number of edited images to generate
+* `size`: string (optional) - Output image dimensions
+* `response_format`: string (default: "b64_json") - Either `url` or `b64_json`
+* `input_fidelity`: string (default: "low") - Either `low` or `high` - Controls how closely the output follows the input image
+* `stream`: string (default: "false") - Enable streaming
+* `partial_images`: string (default: "0") - Number of partial images to stream
+* `quality`: string (default: "medium") - Either `high`, `medium`, or `low`
+* `output_format`: string (default: "png") - Either `png`, `jpeg`, or `webp`
+* `advanced_params`: string (optional) - JSON-encoded advanced parameters
+
 **Response:**
-OpenAI-compatible image generation response.
+Same format as `/v1/images/generations`.
 
 ### Benchmarked Image Generation
 
@@ -223,16 +585,15 @@ Same as `/v1/images/generations`, but also returns generation statistics.
 Same schema as `/v1/images/generations`.
 
 **Response:**
-Image generation plus benchmarking metrics.
+Image generation plus benchmarking metrics including:
 
-### Image Editing
-
-**POST** `/v1/images/edits`
-
-Executes an image editing request using an OpenAI-compatible schema with additional advanced_params (same as `/v1/images/generations`).
-
-**Response:**
-Same format as `/v1/images/generations`.
+* `seconds_per_step` - Average time per denoising step
+* `total_generation_time` - Total generation time
+* `num_inference_steps` - Number of inference steps used
+* `num_images` - Number of images generated
+* `image_width` - Output image width
+* `image_height` - Output image height
+* `peak_memory_usage` - Peak memory used during generation
 
 ### Benchmarked Image Editing
 
@@ -246,36 +607,127 @@ Same schema as `/v1/images/edits`.
 **Response:**
 Same format as `/bench/images/generations`, including `generation_stats`.
 
-## 6. Complete Endpoint Summary
+### List Images
+
+**GET** `/images`
+
+List all stored images.
+
+**Response:**
+Array of image metadata including URLs and expiration times.
+
+### Get Image
+
+**GET** `/images/{image_id}`
+
+Retrieve a stored image by ID.
+
+**Path parameters:**
+
+* `image_id`: string, ID of the image
+
+**Response:**
+Image file with appropriate content type.
+
+## 7. Complete Endpoint Summary
 
 ```
+# General
 GET     /node_id
 GET     /state
 GET     /events
 
+# Instance Management
 POST    /instance
 GET     /instance/{instance_id}
 DELETE  /instance/{instance_id}
-
 GET     /instance/previews
 GET     /instance/placement
 POST    /place_instance
 
+# Models
 GET     /models
 GET     /v1/models
+POST    /models/add
+DELETE  /models/custom/{model_id}
+GET     /models/search
 
+# Text Generation (OpenAI Chat Completions)
 POST    /v1/chat/completions
 POST    /bench/chat/completions
 
+# Text Generation (Claude Messages API)
+POST    /v1/messages
+
+# Text Generation (OpenAI Responses API)
+POST    /v1/responses
+
+# Text Generation (Ollama API)
+POST    /ollama/api/chat
+POST    /ollama/api/api/chat
+POST    /ollama/api/v1/chat
+POST    /ollama/api/generate
+GET     /ollama/api/tags
+GET     /ollama/api/api/tags
+GET     /ollama/api/v1/tags
+POST    /ollama/api/show
+GET     /ollama/api/ps
+GET     /ollama/api/version
+HEAD    /ollama/
+HEAD    /ollama/api/version
+
+# Command Control
+POST    /v1/cancel/{command_id}
+
+# Image Generation
 POST    /v1/images/generations
 POST    /bench/images/generations
 POST    /v1/images/edits
 POST    /bench/images/edits
+GET     /images
+GET     /images/{image_id}
 ```
 
-## 7. Notes
+## 8. Notes
 
-* The `/v1/chat/completions` endpoint is compatible with the OpenAI Chat API format, so existing OpenAI clients can be pointed to EXO by changing the base URL.
-* The `/v1/images/generations` and `/v1/images/edits` endpoints are compatible with the OpenAI Images API format.
-* The instance placement endpoints allow you to plan and preview cluster allocations before actually creating instances.
-* The `/events` and `/state` endpoints are primarily intended for operational visibility and debugging.
+### API Compatibility
+
+EXO provides multiple API-compatible interfaces:
+
+* **OpenAI Chat Completions API** - Compatible with OpenAI clients and tools
+* **Claude Messages API** - Compatible with Anthropic's Claude API format
+* **OpenAI Responses API** - Compatible with OpenAI's Responses API format
+* **Ollama API** - Compatible with Ollama and tools like OpenWebUI
+
+Existing OpenAI, Claude, or Ollama clients can be pointed to EXO by changing the base URL.
+
+### Custom Models
+
+You can add custom models from HuggingFace using the `/models/add` endpoint. Custom models are identified by the `is_custom` field in model list responses.
+
+**Security:** Models requiring `trust_remote_code` must be explicitly enabled (default is false) for security. Only enable this if you trust the model's remote code.
+
+### Usage Statistics
+
+Chat completion responses include usage statistics with:
+
+* `prompt_tokens` - Number of tokens in the prompt
+* `completion_tokens` - Number of tokens generated
+* `total_tokens` - Sum of prompt and completion tokens
+
+### Request Cancellation
+
+You can cancel active requests by:
+
+1. Closing the HTTP connection (for streaming requests)
+2. Calling `/v1/cancel/{command_id}` (for any request)
+
+The server detects cancellation and stops processing immediately.
+
+### Instance Placement
+
+The instance placement endpoints allow you to plan and preview cluster allocations before creating instances. This helps optimize resource usage across nodes.
+
+### Observability
+
+The `/events` and `/state` endpoints are primarily intended for operational visibility and debugging.

docs/architecture.md

@@ -28,6 +28,26 @@ There are currently 5 major systems:
     
     Implements a distributed algorithm for master election in unstable networking conditions
 
+## API Layer
+
+The API system uses multiple adapters to support multiple API formats, converting them to a single request / response type.
+
+### Adapter Pattern
+
+Adapters convert between external API formats and EXO's internal types:
+
+```
+Chat Completions → [adapter] → TextGenerationTaskParams → Application
+Claude Messages  → [adapter] → TextGenerationTaskParams → Application
+Responses API    → [adapter] → TextGenerationTaskParams → Application
+Ollama API       → [adapter] → TextGenerationTaskParams → Application
+```
+
+Each adapter implements two key functions:
+1. **Request conversion**: Converts API-specific requests to `TextGenerationTaskParams`
+2. **Response generation**: Converts internal `TokenChunk` streams back to API-specific formats (streaming and non-streaming)
+
+
 ## Topics
 
 There are currently 5 topics:

flake.nix

@@ -72,9 +72,8 @@
       ];
 
       perSystem =
-        { config, self', inputs', pkgs, lib, system, ... }:
+        { config, self', pkgs, lib, system, ... }:
         let
-          fenixToolchain = inputs'.fenix.packages.complete;
           # Use pinned nixpkgs for swift-format (swift is broken on x86_64-linux in newer nixpkgs)
           pkgsSwift = import inputs.nixpkgs-swift { inherit system; };
         in
@@ -85,6 +84,17 @@
             config.allowUnfreePredicate = pkg: (pkg.pname or "") == "metal-toolchain";
             overlays = [
               (import ./nix/apple-sdk-overlay.nix)
+              (final: prev: {
+                macmon = prev.macmon.overrideAttrs (_: {
+                  version = "git";
+                  src = final.fetchFromGitHub {
+                    owner = "swiftraccoon";
+                    repo = "macmon";
+                    rev = "9154d234f763fbeffdcb4135d0bbbaf80609699b";
+                    hash = "sha256-CwhilKNbs5XL9/tF5DMwyPBlE/hpmjGNTuxQ36sM50M=";
+                  };
+                });
+              })
             ];
           };
           treefmt = {
@@ -109,20 +119,22 @@
                 package = pkgsSwift.swiftPackages.swift-format;
               };
               shfmt.enable = true;
+              taplo.enable = true;
             };
           };
 
           packages = lib.optionalAttrs pkgs.stdenv.hostPlatform.isDarwin (
             let
               uvLock = builtins.fromTOML (builtins.readFile ./uv.lock);
-              mlxPackage = builtins.head (builtins.filter (p: p.name == "mlx") uvLock.package);
+              mlxPackage = builtins.head (builtins.filter (p: p.name == "mlx" && p.source ? git) uvLock.package);
               uvLockMlxVersion = mlxPackage.version;
+              uvLockMlxRev = builtins.elemAt (builtins.split "#" mlxPackage.source.git) 2;
             in
             {
               metal-toolchain = pkgs.callPackage ./nix/metal-toolchain.nix { };
               mlx = pkgs.callPackage ./nix/mlx.nix {
                 inherit (self'.packages) metal-toolchain;
-                inherit uvLockMlxVersion;
+                inherit uvLockMlxVersion uvLockMlxRev;
               };
               default = self'.packages.exo;
             }

justfile

@@ -1,5 +1,8 @@
 export NIX_CONFIG := "extra-experimental-features = nix-command flakes"
 
+default: lint fmt
+all: lint fmt check
+
 fmt:
     treefmt || nix fmt
 
@@ -31,6 +34,10 @@ build-dashboard:
 package:
     uv run pyinstaller packaging/pyinstaller/exo.spec
 
+build-app: package
+    xcodebuild build -project app/EXO/EXO.xcodeproj -scheme EXO -configuration Debug -derivedDataPath app/EXO/build
+    @echo "\nBuild complete. Run with:\n  open {{justfile_directory()}}/app/EXO/build/Build/Products/Debug/EXO.app"
+
 clean:
     rm -rf **/__pycache__
     rm -rf target/

nix/mlx.nix

@@ -11,6 +11,7 @@
 , fmt
 , python313Packages
 , uvLockMlxVersion
+, uvLockMlxRev
 }:
 
 assert stdenv.isDarwin;
@@ -41,16 +42,14 @@ let
 
   mlx = stdenv.mkDerivation rec {
     pname = "mlx";
-    version = let v = "0.30.6"; in
-      assert v == uvLockMlxVersion || throw "MLX version mismatch: nix/mlx.nix has ${v} but uv.lock has ${uvLockMlxVersion}. Update both the version and hash in nix/mlx.nix.";
-      v;
+    version = uvLockMlxVersion;
     pyproject = true;
 
     src = fetchFromGitHub {
-      owner = "ml-explore";
-      repo = "mlx";
-      tag = "v${version}";
-      hash = "sha256-avD5EGhwgmPdXLAyQSqTO6AXk/W3ziH+f6AetjK3Sdo=";
+      owner = "rltakashige";
+      repo = "mlx-jaccl-fix-small-recv";
+      rev = uvLockMlxRev;
+      hash = "sha256-GosFIWxIB48Egb1MqJrR3xhsUsQeWdRk5rV93USY6wQ=";
     };
 
     patches = [

packaging/dmg/create-dmg.sh

@@ -0,0 +1,112 @@
+#!/usr/bin/env bash
+# create-dmg.sh — Build a polished macOS DMG installer for EXO
+#
+# Usage:
+#   ./packaging/dmg/create-dmg.sh <app-path> <output-dmg> [volume-name]
+#
+# Example:
+#   ./packaging/dmg/create-dmg.sh output/EXO.app EXO-1.0.0.dmg "EXO"
+#
+# Creates a DMG with:
+#   - Custom background image with drag-to-Applications arrow
+#   - App icon on left, Applications alias on right
+#   - Proper window size and icon positioning
+set -euo pipefail
+
+APP_PATH="${1:?Usage: create-dmg.sh <app-path> <output-dmg> [volume-name]}"
+OUTPUT_DMG="${2:?Usage: create-dmg.sh <app-path> <output-dmg> [volume-name]}"
+VOLUME_NAME="${3:-EXO}"
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+BACKGROUND_SCRIPT="${SCRIPT_DIR}/generate-background.py"
+TEMP_DIR="$(mktemp -d)"
+DMG_STAGING="${TEMP_DIR}/dmg-root"
+TEMP_DMG="${TEMP_DIR}/temp.dmg"
+BACKGROUND_PNG="${TEMP_DIR}/background.png"
+
+cleanup() { rm -rf "$TEMP_DIR"; }
+trap cleanup EXIT
+
+echo "==> Creating DMG installer for ${VOLUME_NAME}"
+
+# ── Step 1: Generate background image ────────────────────────────────────────
+if command -v python3 &>/dev/null; then
+  python3 "$BACKGROUND_SCRIPT" "$BACKGROUND_PNG"
+  echo "    Background image generated"
+else
+  echo "    Warning: python3 not found, skipping custom background"
+  BACKGROUND_PNG=""
+fi
+
+# ── Step 2: Prepare staging directory ─────────────────────────────────────────
+mkdir -p "$DMG_STAGING"
+cp -R "$APP_PATH" "$DMG_STAGING/"
+ln -s /Applications "$DMG_STAGING/Applications"
+
+# ── Step 3: Create writable DMG ──────────────────────────────────────────────
+# Calculate required size (app size + 20MB headroom)
+APP_SIZE_KB=$(du -sk "$APP_PATH" | cut -f1)
+DMG_SIZE_KB=$((APP_SIZE_KB + 20480))
+
+hdiutil create \
+  -volname "$VOLUME_NAME" \
+  -size "${DMG_SIZE_KB}k" \
+  -fs HFS+ \
+  -layout SPUD \
+  "$TEMP_DMG"
+
+# ── Step 4: Mount and configure ──────────────────────────────────────────────
+MOUNT_DIR=$(hdiutil attach "$TEMP_DMG" -readwrite -noverify | awk -F'\t' '/Apple_HFS/ {gsub(/^[[:space:]]+|[[:space:]]+$/, "", $NF); print $NF}')
+echo "    Mounted at: $MOUNT_DIR"
+
+# Copy contents
+cp -R "$DMG_STAGING/"* "$MOUNT_DIR/"
+
+# Add background image
+if [[ -n $BACKGROUND_PNG && -f $BACKGROUND_PNG ]]; then
+  mkdir -p "$MOUNT_DIR/.background"
+  cp "$BACKGROUND_PNG" "$MOUNT_DIR/.background/background.png"
+fi
+
+# ── Step 5: Configure window appearance via AppleScript ──────────────────────
+# Window: 800×400, app icon on left, Applications on right (matches Ollama layout)
+# Background image is 1600×740 (2× retina for 800×400 logical window).
+APP_NAME="$(basename "$APP_PATH")"
+
+osascript <<APPLESCRIPT
+tell application "Finder"
+    tell disk "$VOLUME_NAME"
+        open
+        set current view of container window to icon view
+        set toolbar visible of container window to false
+        set statusbar visible of container window to false
+        set bounds of container window to {200, 120, 1000, 520}
+        set opts to icon view options of container window
+        set icon size of opts to 128
+        set text size of opts to 12
+        set arrangement of opts to not arranged
+        if exists file ".background:background.png" then
+            set background picture of opts to file ".background:background.png"
+        end if
+        set position of item "$APP_NAME" of container window to {200, 190}
+        set position of item "Applications" of container window to {600, 190}
+        close
+        open
+        update without registering applications
+        delay 1
+        close
+    end tell
+end tell
+APPLESCRIPT
+
+echo "    Window layout configured"
+
+# Ensure Finder updates are flushed
+sync
+
+# ── Step 6: Finalise ─────────────────────────────────────────────────────────
+hdiutil detach "$MOUNT_DIR" -quiet
+hdiutil convert "$TEMP_DMG" -format UDZO -imagekey zlib-level=9 -o "$OUTPUT_DMG"
+
+echo "==> DMG created: $OUTPUT_DMG"
+echo "    Size: $(du -h "$OUTPUT_DMG" | cut -f1)"

packaging/dmg/generate-background.py

@@ -0,0 +1,91 @@
+#!/usr/bin/env python3
+"""Generate the DMG background image with a centered drag-to-Applications arrow.
+
+The output is a 1600×740 retina PNG (2× for 800×400 logical window).
+Icons are positioned at (200, 190) and (600, 190) in logical coordinates;
+the arrow is drawn centered between them.
+
+Usage:
+    python3 generate-background.py [output.png]
+
+If no output path is given, overwrites the bundled background.png in-place.
+"""
+
+from __future__ import annotations
+
+import math
+import sys
+from pathlib import Path
+
+from PIL import Image, ImageDraw
+
+# Retina dimensions (2× logical 800×400)
+WIDTH = 1600
+HEIGHT = 740
+
+# Icon positions in logical coords → retina coords
+# App icon at (200, 190), Applications at (600, 190)
+APP_X = 200 * 2  # 400
+APPS_X = 600 * 2  # 1200
+ICON_Y = 190 * 2  # 380
+
+# Arrow drawn between icons, slightly above icon center
+ARROW_START_X = APP_X + 160  # past the icon
+ARROW_END_X = APPS_X - 160  # before the Applications icon
+ARROW_Y = ICON_Y  # same height as icons
+ARROW_RISE = 120  # upward arc height
+
+
+def draw_arrow(draw: ImageDraw.ImageDraw) -> None:
+    """Draw a hand-drawn-style curved arrow from app icon toward Applications."""
+    color = (30, 30, 30)
+    line_width = 8
+
+    # Compute bezier curve points for a gentle upward arc
+    points: list[tuple[float, float]] = []
+    steps = 80
+    for i in range(steps + 1):
+        t = i / steps
+        # Quadratic bezier: start → control → end
+        cx = (ARROW_START_X + ARROW_END_X) / 2
+        cy = ARROW_Y - ARROW_RISE
+        x = (1 - t) ** 2 * ARROW_START_X + 2 * (1 - t) * t * cx + t**2 * ARROW_END_X
+        y = (1 - t) ** 2 * ARROW_Y + 2 * (1 - t) * t * cy + t**2 * ARROW_Y
+        points.append((x, y))
+
+    # Draw the curve as connected line segments
+    for i in range(len(points) - 1):
+        draw.line([points[i], points[i + 1]], fill=color, width=line_width)
+
+    # Arrowhead at the end
+    end_x, end_y = points[-1]
+    # Direction from second-to-last to last point
+    prev_x, prev_y = points[-3]
+    angle = math.atan2(end_y - prev_y, end_x - prev_x)
+    head_len = 36
+    head_angle = math.radians(25)
+
+    left_x = end_x - head_len * math.cos(angle - head_angle)
+    left_y = end_y - head_len * math.sin(angle - head_angle)
+    right_x = end_x - head_len * math.cos(angle + head_angle)
+    right_y = end_y - head_len * math.sin(angle + head_angle)
+
+    draw.polygon(
+        [(end_x, end_y), (left_x, left_y), (right_x, right_y)],
+        fill=color,
+    )
+
+
+def generate_background(output_path: str) -> None:
+    """Generate a white DMG background with a centered arrow."""
+    img = Image.new("RGBA", (WIDTH, HEIGHT), (255, 255, 255, 255))
+    draw = ImageDraw.Draw(img)
+    draw_arrow(draw)
+    img.save(output_path, "PNG")
+
+
+if __name__ == "__main__":
+    default_output = str(Path(__file__).parent / "background.png")
+    out = sys.argv[1] if len(sys.argv) >= 2 else default_output
+    generate_background(out)
+    print(f"Background image written to {out}")

packaging/pyinstaller/exo.spec

@@ -71,7 +71,9 @@ MACMON_PATH = shutil.which("macmon")
 if MACMON_PATH is None:
     raise SystemExit(
         "macmon binary not found in PATH. "
-        "Install it via: brew install macmon"
+        "Install the pinned fork used by exo via: "
+        "cargo install --git https://github.com/swiftraccoon/macmon "
+        "--rev 9154d234f763fbeffdcb4135d0bbbaf80609699b macmon --force"
     )
 
 BINARIES: list[tuple[str, str]] = [
@@ -120,4 +122,3 @@ coll = COLLECT(
     upx_exclude=[],
     name="exo",
 )
-

pyproject.toml

@@ -1,35 +1,34 @@
 [project]
 name = "exo"
-version = "0.3.0"
+version = "0.3.69"
 description = "Exo"
 readme = "README.md"
 requires-python = ">=3.13"
 dependencies = [
-    "aiofiles>=24.1.0",
-    "aiohttp>=3.12.14",
-    "types-aiofiles>=24.1.0.20250708",
-    "pydantic>=2.11.7",
-    "fastapi>=0.116.1",
-    "filelock>=3.18.0",
-    "rustworkx>=0.17.1",
-    "huggingface-hub>=0.33.4",
-    "psutil>=7.0.0",
-    "loguru>=0.7.3",
-    "exo_pyo3_bindings", # rust bindings
-    "anyio==4.11.0",
-    "mlx==0.30.6; sys_platform == 'darwin'",
-    "mlx[cpu]==0.30.6; sys_platform == 'linux'",
-    "mlx-lm==0.30.6",
-    "tiktoken>=0.12.0", # required for kimi k2 tokenizer
-    "hypercorn>=0.18.0",
-    "openai-harmony>=0.0.8",
-    "httpx>=0.28.1",
-    "tomlkit>=0.14.0",
-    "pillow>=11.0,<12.0", # compatibility with mflux
-    "mflux==0.15.5",
-    "python-multipart>=0.0.21",
-    "msgspec>=0.19.0",
-    "zstandard>=0.23.0",
+  "aiofiles>=24.1.0",
+  "aiohttp>=3.12.14",
+  "types-aiofiles>=24.1.0.20250708",
+  "pydantic>=2.11.7",
+  "fastapi>=0.116.1",
+  "filelock>=3.18.0",
+  "rustworkx>=0.17.1",
+  "huggingface-hub>=0.33.4",
+  "psutil>=7.0.0",
+  "loguru>=0.7.3",
+  "exo_pyo3_bindings",                         # rust bindings
+  "anyio==4.11.0",
+  "mlx; sys_platform == 'darwin'",
+  "mlx[cpu]==0.30.6; sys_platform == 'linux'",
+  "mlx-lm",
+  "tiktoken>=0.12.0",                          # required for kimi k2 tokenizer
+  "hypercorn>=0.18.0",
+  "openai-harmony>=0.0.8",
+  "httpx>=0.28.1",
+  "tomlkit>=0.14.0",
+  "mflux==0.17.2",
+  "python-multipart>=0.0.21",
+  "msgspec>=0.19.0",
+  "zstandard>=0.23.0",
 ]
 
 [project.scripts]
@@ -38,12 +37,12 @@ exo = "exo.main:main"
 # dependencies only required for development
 [dependency-groups]
 dev = [
-    "basedpyright>=1.29.0",
-    "pyinstaller>=6.17.0",
-    "pytest>=8.4.0",
-    "pytest-asyncio>=1.0.0",
-    "pytest-env",
-    "ruff>=0.11.13",
+  "basedpyright>=1.29.0",
+  "pyinstaller>=6.17.0",
+  "pytest>=8.4.0",
+  "pytest-asyncio>=1.0.0",
+  "pytest-env",
+  "ruff>=0.11.13",
 ]
 
 # mlx[cuda] requires a newer version of mlx. the ideal on linux is: default to mlx[cpu] unless[cuda] specified.
@@ -57,14 +56,12 @@ dev = [
 ###
 
 [tool.uv.workspace]
-members = [
-    "rust/exo_pyo3_bindings",
-    "bench",
-]
+members = ["rust/exo_pyo3_bindings", "bench"]
 
 [tool.uv.sources]
 exo_pyo3_bindings = { workspace = true }
-#mlx-lm = { git = "https://github.com/davidmcc73/mlx-lm", branch = "stable" }
+mlx = { git = "https://github.com/rltakashige/mlx-jaccl-fix-small-recv.git", branch = "address-rdma-gpu-locks", marker = "sys_platform == 'darwin'" }
+mlx-lm = { git = "https://github.com/rltakashige/mlx-lm", branch = "leo/fix-deepseek-v32-indexer" }
 # Uncomment to use local mlx/mlx-lm development versions:
 # mlx = { path = "/Users/Shared/mlx", editable=true }
 # mlx-lm = { path = "/Users/Shared/mlx-lm", editable=true }
@@ -94,7 +91,15 @@ reportUnnecessaryTypeIgnoreComment = "error"
 pythonVersion = "3.13"
 pythonPlatform = "Darwin"
 
-exclude = ["**/.venv", "**/venv", "**/__pycache__", "**/exo_scripts", "**/.direnv", "**/rust", "**/.github"]
+exclude = [
+  "**/.venv",
+  "**/venv",
+  "**/__pycache__",
+  "**/exo_scripts",
+  "**/.direnv",
+  "**/rust",
+  "**/.github",
+]
 stubPath = ".mlx_typings"
 
 [[tool.basedpyright.executionEnvironments]]
@@ -108,17 +113,19 @@ root = "src"
 [tool.uv]
 required-version = ">=0.8.6"
 prerelease = "allow"
-environments = [
-    "sys_platform == 'darwin'",
-    "sys_platform == 'linux'",
-]
+environments = ["sys_platform == 'darwin'", "sys_platform == 'linux'"]
 
 ###
 # ruff configuration
 ###
 
 [tool.ruff]
-extend-exclude = ["shared/protobufs/**", "*mlx_typings/**", "rust/exo_pyo3_bindings/**"]
+extend-exclude = [
+  "shared/protobufs/**",
+  "*mlx_typings/**",
+  "rust/exo_pyo3_bindings/**",
+  "bench/vendor/**",
+]
 
 [tool.ruff.lint]
 extend-select = ["I", "N", "B", "A", "PIE", "SIM"]
@@ -126,13 +133,7 @@ extend-select = ["I", "N", "B", "A", "PIE", "SIM"]
 [tool.pytest.ini_options]
 pythonpath = "."
 asyncio_mode = "auto"
-markers = [
-    "slow: marks tests as slow (deselected by default)"
-]
-env = [
-  "EXO_TESTS=1"
-]
-addopts = "-m 'not slow'"
-filterwarnings = [
-    "ignore:builtin type Swig:DeprecationWarning",
-]
+markers = ["slow: marks tests as slow (deselected by default)"]
+env = ["EXO_TESTS=1"]
+addopts = "-m 'not slow' --ignore=tests/start_distributed_test.py"
+filterwarnings = ["ignore:builtin type Swig:DeprecationWarning"]

python/parts.nix

@@ -43,6 +43,27 @@
             final.setuptools
           ];
         });
+        # rouge-score and sacrebleu don't declare setuptools as a build dependency
+        rouge-score = prev.rouge-score.overrideAttrs (old: {
+          nativeBuildInputs = (old.nativeBuildInputs or [ ]) ++ [
+            final.setuptools
+          ];
+        });
+        sacrebleu = prev.sacrebleu.overrideAttrs (old: {
+          nativeBuildInputs = (old.nativeBuildInputs or [ ]) ++ [
+            final.setuptools
+          ];
+        });
+        sqlitedict = prev.sqlitedict.overrideAttrs (old: {
+          nativeBuildInputs = (old.nativeBuildInputs or [ ]) ++ [
+            final.setuptools
+          ];
+        });
+        word2number = prev.word2number.overrideAttrs (old: {
+          nativeBuildInputs = (old.nativeBuildInputs or [ ]) ++ [
+            final.setuptools
+          ];
+        });
       } // lib.optionalAttrs pkgs.stdenv.hostPlatform.isDarwin {
         # Use our pure Nix-built MLX with Metal support (macOS only)
         mlx = self'.packages.mlx;
@@ -58,6 +79,21 @@
         lib.optionalAttrs pkgs.stdenv.hostPlatform.isLinux (
           (lib.mapAttrs (_: ignoreMissing) nvidiaPackages) // {
             mlx = ignoreMissing prev.mlx;
+            mlx-cuda-13 = prev.mlx-cuda-13.overrideAttrs (old: {
+              buildInputs = (old.buildInputs or [ ]) ++ [
+                final.nvidia-cublas
+                final.nvidia-cuda-nvrtc
+                final.nvidia-cudnn-cu13
+                final.nvidia-nccl-cu13
+              ];
+              preFixup = ''
+                addAutoPatchelfSearchPath ${final.nvidia-cublas}
+                addAutoPatchelfSearchPath ${final.nvidia-cuda-nvrtc}
+                addAutoPatchelfSearchPath ${final.nvidia-cudnn-cu13}
+                addAutoPatchelfSearchPath ${final.nvidia-nccl-cu13}
+              '';
+              autoPatchelfIgnoreMissingDeps = [ "libcuda.so.1" ];
+            });
             torch = ignoreMissing prev.torch;
             triton = ignoreMissing prev.triton;
           }
@@ -74,14 +110,28 @@
           linuxOverlay
         ]
       );
-      exoVenv = pythonSet.mkVirtualEnv "exo-env" workspace.deps.default;
+      # mlx-cpu and mlx-cuda-13 both ship mlx/ site-packages files; keep first.
+      # mlx-cpu/mlx-cuda-13 and nvidia-cudnn-cu12/cu13 ship overlapping files.
+      venvCollisionPaths = lib.optionals pkgs.stdenv.hostPlatform.isLinux [
+        "lib/python3.13/site-packages/mlx*"
+        "lib/python3.13/site-packages/nvidia*"
+      ];
+
+      # Exclude bench deps from main env (bench has its own benchVenv)
+      exoDeps = removeAttrs workspace.deps.default [ "exo-bench" ];
+
+      exoVenv = (pythonSet.mkVirtualEnv "exo-env" exoDeps).overrideAttrs {
+        venvIgnoreCollisions = venvCollisionPaths;
+      };
 
       # Virtual environment with dev dependencies for testing
-      testVenv = pythonSet.mkVirtualEnv "exo-test-env" (
-        workspace.deps.default // {
+      testVenv = (pythonSet.mkVirtualEnv "exo-test-env" (
+        exoDeps // {
           exo = [ "dev" ]; # Include pytest, pytest-asyncio, pytest-env
         }
-      );
+      )).overrideAttrs {
+        venvIgnoreCollisions = venvCollisionPaths;
+      };
 
       mkPythonScript = name: path: pkgs.writeShellApplication {
         inherit name;
@@ -132,6 +182,8 @@
           exo-test-env = testVenv;
         } // {
         exo-bench = mkBenchScript "exo-bench" (inputs.self + /bench/exo_bench.py);
+        exo-eval = mkBenchScript "exo-eval" (inputs.self + /bench/exo_eval.py);
+        exo-eval-tool-calls = mkBenchScript "exo-eval-tool-calls" (inputs.self + /bench/eval_tool_calls.py);
         exo-get-all-models-on-cluster = mkSimplePythonScript "exo-get-all-models-on-cluster" (inputs.self + /tests/get_all_models_on_cluster.py);
       };
 

resources/inference_model_cards/mlx-community--DeepSeek-V3.1-4bit.toml

@@ -1,12 +1,13 @@
 model_id = "mlx-community/DeepSeek-V3.1-4bit"
 n_layers = 61
 hidden_size = 7168
+num_key_value_heads = 128
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "deepseek"
 quantization = "4bit"
 base_model = "DeepSeek V3.1"
-capabilities = ["text", "thinking"]
+capabilities = ["text", "thinking", "thinking_toggle"]
 
 [storage_size]
 in_bytes = 405874409472

resources/inference_model_cards/mlx-community--DeepSeek-V3.1-8bit.toml

@@ -1,12 +1,13 @@
 model_id = "mlx-community/DeepSeek-V3.1-8bit"
 n_layers = 61
 hidden_size = 7168
+num_key_value_heads = 128
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "deepseek"
 quantization = "8bit"
 base_model = "DeepSeek V3.1"
-capabilities = ["text", "thinking"]
+capabilities = ["text", "thinking", "thinking_toggle"]
 
 [storage_size]
 in_bytes = 765577920512

resources/inference_model_cards/mlx-community--DeepSeek-V3.2-4bit.toml

@@ -0,0 +1,13 @@
+model_id = "mlx-community/DeepSeek-V3.2-4bit"
+n_layers = 61
+hidden_size = 7168
+num_key_value_heads = 128
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "deepseek"
+quantization = "4bit"
+base_model = "DeepSeek V3.2"
+capabilities = ["text", "thinking", "thinking_toggle"]
+
+[storage_size]
+in_bytes = 378086226621

resources/inference_model_cards/mlx-community--DeepSeek-V3.2-8bit.toml

@@ -0,0 +1,13 @@
+model_id = "mlx-community/DeepSeek-V3.2-8bit"
+n_layers = 61
+hidden_size = 7168
+num_key_value_heads = 128
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "deepseek"
+quantization = "8bit"
+base_model = "DeepSeek V3.2"
+capabilities = ["text", "thinking", "thinking_toggle"]
+
+[storage_size]
+in_bytes = 755957120916

resources/inference_model_cards/mlx-community--GLM-4.5-Air-8bit.toml

@@ -1,12 +1,13 @@
 model_id = "mlx-community/GLM-4.5-Air-8bit"
 n_layers = 46
 hidden_size = 4096
+num_key_value_heads = 8
 supports_tensor = false
 tasks = ["TextGeneration"]
 family = "glm"
 quantization = "8bit"
 base_model = "GLM 4.5 Air"
-capabilities = ["text", "thinking"]
+capabilities = ["text", "thinking", "thinking_toggle"]
 
 [storage_size]
 in_bytes = 122406567936

resources/inference_model_cards/mlx-community--GLM-4.5-Air-bf16.toml

@@ -1,12 +1,13 @@
 model_id = "mlx-community/GLM-4.5-Air-bf16"
 n_layers = 46
 hidden_size = 4096
+num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"
 quantization = "bf16"
 base_model = "GLM 4.5 Air"
-capabilities = ["text", "thinking"]
+capabilities = ["text", "thinking", "thinking_toggle"]
 
 [storage_size]
 in_bytes = 229780750336

resources/inference_model_cards/mlx-community--GLM-4.7-4bit.toml

@@ -1,12 +1,13 @@
 model_id = "mlx-community/GLM-4.7-4bit"
 n_layers = 91
 hidden_size = 5120
+num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"
 quantization = "4bit"
 base_model = "GLM 4.7"
-capabilities = ["text", "thinking"]
+capabilities = ["text", "thinking", "thinking_toggle"]
 
 [storage_size]
 in_bytes = 198556925568

resources/inference_model_cards/mlx-community--GLM-4.7-6bit.toml

@@ -1,12 +1,13 @@
 model_id = "mlx-community/GLM-4.7-6bit"
 n_layers = 91
 hidden_size = 5120
+num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"
 quantization = "6bit"
 base_model = "GLM 4.7"
-capabilities = ["text", "thinking"]
+capabilities = ["text", "thinking", "thinking_toggle"]
 
 [storage_size]
 in_bytes = 286737579648

resources/inference_model_cards/mlx-community--GLM-4.7-8bit-gs32.toml

@@ -1,12 +1,13 @@
 model_id = "mlx-community/GLM-4.7-8bit-gs32"
 n_layers = 91
 hidden_size = 5120
+num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"
 quantization = "8bit"
 base_model = "GLM 4.7"
-capabilities = ["text", "thinking"]
+capabilities = ["text", "thinking", "thinking_toggle"]
 
 [storage_size]
 in_bytes = 396963397248

resources/inference_model_cards/mlx-community--GLM-4.7-Flash-4bit.toml

@@ -1,12 +1,13 @@
 model_id = "mlx-community/GLM-4.7-Flash-4bit"
 n_layers = 47
 hidden_size = 2048
+num_key_value_heads = 20
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"
 quantization = "4bit"
 base_model = "GLM 4.7 Flash"
-capabilities = ["text", "thinking"]
+capabilities = ["text", "thinking", "thinking_toggle"]
 
 [storage_size]
 in_bytes = 19327352832

resources/inference_model_cards/mlx-community--GLM-4.7-Flash-5bit.toml

@@ -1,12 +1,13 @@
 model_id = "mlx-community/GLM-4.7-Flash-5bit"
 n_layers = 47
 hidden_size = 2048
+num_key_value_heads = 20
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"
 quantization = "5bit"
 base_model = "GLM 4.7 Flash"
-capabilities = ["text", "thinking"]
+capabilities = ["text", "thinking", "thinking_toggle"]
 
 [storage_size]
 in_bytes = 22548578304

resources/inference_model_cards/mlx-community--GLM-4.7-Flash-6bit.toml

@@ -1,12 +1,13 @@
 model_id = "mlx-community/GLM-4.7-Flash-6bit"
 n_layers = 47
 hidden_size = 2048
+num_key_value_heads = 20
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"
 quantization = "6bit"
 base_model = "GLM 4.7 Flash"
-capabilities = ["text", "thinking"]
+capabilities = ["text", "thinking", "thinking_toggle"]
 
 [storage_size]
 in_bytes = 26843545600

resources/inference_model_cards/mlx-community--GLM-4.7-Flash-8bit.toml

@@ -1,12 +1,13 @@
 model_id = "mlx-community/GLM-4.7-Flash-8bit"
 n_layers = 47
 hidden_size = 2048
+num_key_value_heads = 20
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"
 quantization = "8bit"
 base_model = "GLM 4.7 Flash"
-capabilities = ["text", "thinking"]
+capabilities = ["text", "thinking", "thinking_toggle"]
 
 [storage_size]
 in_bytes = 34359738368

resources/inference_model_cards/mlx-community--GLM-5-8bit.toml

@@ -0,0 +1,13 @@
+model_id = "mlx-community/GLM-5-8bit-MXFP8"
+n_layers = 78
+hidden_size = 6144
+num_key_value_heads = 64
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "glm"
+quantization = "8bit"
+base_model = "GLM-5"
+capabilities = ["text", "thinking"]
+
+[storage_size]
+in_bytes = 790517400864

resources/inference_model_cards/mlx-community--GLM-5-MXFP4-Q8.toml

@@ -0,0 +1,13 @@
+model_id = "mlx-community/GLM-5-MXFP4-Q8"
+n_layers = 78
+hidden_size = 6144
+num_key_value_heads = 64
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "glm"
+quantization = "MXFP4-Q8"
+base_model = "GLM-5"
+capabilities = ["text", "thinking"]
+
+[storage_size]
+in_bytes = 405478939008

resources/inference_model_cards/mlx-community--GLM-5-bf16.toml

@@ -0,0 +1,13 @@
+model_id = "mlx-community/GLM-5"
+n_layers = 78
+hidden_size = 6144
+num_key_value_heads = 64
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "glm"
+quantization = "bf16"
+base_model = "GLM-5"
+capabilities = ["text", "thinking"]
+
+[storage_size]
+in_bytes = 1487822475264

resources/inference_model_cards/mlx-community--Kimi-K2-Instruct-4bit.toml

@@ -1,6 +1,7 @@
 model_id = "mlx-community/Kimi-K2-Instruct-4bit"
 n_layers = 61
 hidden_size = 7168
+num_key_value_heads = 64
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "kimi"

resources/inference_model_cards/mlx-community--Kimi-K2-Thinking.toml

@@ -1,12 +1,13 @@
 model_id = "mlx-community/Kimi-K2-Thinking"
 n_layers = 61
 hidden_size = 7168
+num_key_value_heads = 64
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "kimi"
 quantization = ""
 base_model = "Kimi K2"
-capabilities = ["text", "thinking"]
+capabilities = ["text", "thinking", "thinking_toggle"]
 
 [storage_size]
 in_bytes = 706522120192

resources/inference_model_cards/mlx-community--Kimi-K2.5.toml

@@ -1,12 +1,13 @@
 model_id = "mlx-community/Kimi-K2.5"
 n_layers = 61
 hidden_size = 7168
+num_key_value_heads = 64
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "kimi"
 quantization = ""
 base_model = "Kimi K2.5"
-capabilities = ["text", "thinking"]
+capabilities = ["text", "thinking", "thinking_toggle"]
 
 [storage_size]
 in_bytes = 662498705408