Merge branch 'main' into evoisin/fix-prohibited-files-send

CHANGELOG.md

@@ -14,6 +14,11 @@ and this project adheres to
 - 🐛(front) fix target blank links in chat #103
 - 🚑️(posthog) pass str instead of UUID for user PK #134
 - ⚡️(web-search) keep running when tool call fails #137
+- ✨(summarize): new summarize tool integration #78
+
+### Removed
+
+- 🔥(posthog) remove posthog middleware for async mode fix #146
 
 
 ## [0.0.7] - 2025-10-28

src/backend/chat/agents/summarize.py

@@ -4,11 +4,6 @@ import dataclasses
 import logging
 
 from django.conf import settings
-from django.core.files.storage import default_storage
-
-from asgiref.sync import sync_to_async
-from pydantic_ai import RunContext
-from pydantic_ai.messages import ToolReturn
 
 from .base import BaseAgent
 
@@ -26,60 +21,3 @@ class SummarizationAgent(BaseAgent):
             output_type=str,
             **kwargs,
         )
-
-
-@sync_to_async
-def read_document_content(doc):
-    """Read document content asynchronously."""
-    with default_storage.open(doc.key) as f:
-        return doc.file_name, f.read().decode("utf-8")
-
-
-async def hand_off_to_summarization_agent(ctx: RunContext) -> ToolReturn:
-    """
-    Generate a complete, ready-to-use summary of the documents in context
-    (do not request the documents to the user).
-    Return this summary directly to the user WITHOUT any modification,
-    or additional summarization.
-    The summary is already optimized and MUST be presented as-is in the final response
-    or translated preserving the information.
-    """
-    summarization_agent = SummarizationAgent()
-
-    prompt = (
-        "Do not mention the user request in your answer.\n"
-        "User request:\n"
-        "{user_prompt}\n\n"
-        "Document contents:\n"
-        "{documents_prompt}\n"
-    )
-    text_attachment = await sync_to_async(list)(
-        ctx.deps.conversation.attachments.filter(
-            content_type__startswith="text/",
-        )
-    )
-
-    documents = [await read_document_content(doc) for doc in text_attachment]
-
-    documents_prompt = "\n\n".join(
-        [
-            (f"<document>\n<name>\n{name}\n</name>\n<content>\n{content}\n</content>\n</document>")
-            for name, content in documents
-        ]
-    )
-
-    formatted_prompt = prompt.format(
-        user_prompt=ctx.prompt,
-        documents_prompt=documents_prompt,
-    )
-
-    logger.debug("Summarize prompt: %s", formatted_prompt)
-
-    response = await summarization_agent.run(formatted_prompt, usage=ctx.usage)
-
-    logger.debug("Summarize response: %s", response)
-
-    return ToolReturn(
-        return_value=response.output,
-        metadata={"sources": {doc[0] for doc in documents}},
-    )

src/backend/chat/clients/pydantic_ai.py

@@ -7,6 +7,7 @@ changes are needed in views.py or tests.
 """
 
 import dataclasses
+import functools
 import json
 import logging
 import time
@@ -25,7 +26,7 @@ from django.utils.module_loading import import_string
 
 from asgiref.sync import sync_to_async
 from langfuse import get_client
-from pydantic_ai import Agent
+from pydantic_ai import Agent, RunContext
 from pydantic_ai.messages import (
     BinaryContent,
     DocumentUrl,
@@ -59,7 +60,6 @@ from chat.agents.local_media_url_processors import (
     update_history_local_urls,
     update_local_urls,
 )
-from chat.agents.summarize import hand_off_to_summarization_agent
 from chat.ai_sdk_types import (
     LanguageModelV1Source,
     SourceUIPart,
@@ -73,6 +73,7 @@ from chat.clients.pydantic_ui_message_converter import (
 )
 from chat.mcp_servers import get_mcp_servers
 from chat.tools.document_search_rag import add_document_rag_search_tool
+from chat.tools.document_summarize import document_summarize
 from chat.vercel_ai_sdk.core import events_v4, events_v5
 from chat.vercel_ai_sdk.encoder import EventEncoder
 
@@ -493,13 +494,20 @@ class AIAgentService:  # pylint: disable=too-many-instance-attributes
                     "You may add a follow-up question after the summary if needed."
                 )
 
-            @self.conversation_agent.tool
-            async def summarize(ctx) -> ToolReturn:
-                """
-                Summarize the documents for the user, only when asked for,
-                the documents are in my context.
-                """
-                return await hand_off_to_summarization_agent(ctx)
+            # Inform the model (system-level) that documents are attached and available
+            @self.conversation_agent.system_prompt
+            def attached_documents_note() -> str:
+                return (
+                    "[Internal context] User documents are attached to this conversation. "
+                    "Do not request re-upload of documents; consider them already available "
+                    "via the internal store."
+                )
+
+            @self.conversation_agent.tool(name="summarize", retries=2)
+            @functools.wraps(document_summarize)
+            async def summarize(ctx: RunContext, *args, **kwargs) -> ToolReturn:
+                """Wrap the document_summarize tool to provide context and add the tool."""
+                return await document_summarize(ctx, *args, **kwargs)
         else:
             conversation_documents = [
                 cd

src/backend/chat/tests/conftest.py

@@ -8,6 +8,7 @@ from django.utils import formats, timezone
 
 import pytest
 
+from chat.agents.summarize import SummarizationAgent
 from chat.clients.pydantic_ai import AIAgentService
 
 logger = logging.getLogger(__name__)
@@ -50,6 +51,40 @@ def mock_ai_agent_service_fixture():
     yield _mock_service
 
 
+@pytest.fixture(name="mock_summarization_agent")
+def mock_summarization_agent_fixture():
+    """Fixture to mock SummarizationAgent with a custom model."""
+
+    @contextmanager
+    def _mock_agent(model):
+        """Context manager to mock SummarizationAgent with a custom model."""
+        with ExitStack() as stack:
+
+            class SummarizationAgentMock(SummarizationAgent):
+                """Mocked SummarizationAgent to override the model."""
+
+                def __init__(self, **kwargs):
+                    super().__init__(**kwargs)
+                    # We cannot use stack.enter_context(agent.override(model=model))
+                    # Because the agent is used outside of this context manager.
+                    # So we directly override the protected member.
+                    logger.info("Overriding SummarizationAgent model with %s", model)
+                    self._model = model  # pylint: disable=protected-access
+
+            # Mock the SummarizationAgent in all relevant modules, because first import wins
+            stack.enter_context(
+                patch("chat.agents.summarize.SummarizationAgent", new=SummarizationAgentMock)
+            )
+            stack.enter_context(
+                patch(
+                    "chat.tools.document_summarize.SummarizationAgent", new=SummarizationAgentMock
+                )
+            )
+            yield
+
+    yield _mock_agent
+
+
 PIXEL_PNG = (
     b"\x89PNG\r\n\x1a\n\x00\x00\x00\rIHDR\x00\x00\x00\x01\x00\x00\x00\x01\x08\x06\x00"
     b"\x00\x00\x1f\x15\xc4\x89\x00\x00\x00\nIDATx\x9cc\xf8\xff\xff?\x00\x05\xfe\x02\xfe"

src/backend/chat/tests/tools/test_document_summarize.py

@@ -0,0 +1,472 @@
+"""Tests for document_summarize functionality."""
+
+import io
+from unittest import mock
+
+from django.core.files.storage import default_storage
+
+import pytest
+from pydantic_ai import ModelResponse, RunContext, TextPart
+from pydantic_ai.exceptions import ModelRetry
+from pydantic_ai.models.function import FunctionModel
+from pydantic_ai.usage import RunUsage
+
+from chat.agents.summarize import SummarizationAgent
+from chat.llm_configuration import LLModel, LLMProvider
+from chat.tools.document_summarize import document_summarize, summarize_chunk
+
+
+@pytest.fixture(autouse=True)
+def fixture_summarization_agent_config(settings):
+    """Fixture to set used settings for agent configuration."""
+    settings.LLM_CONFIGURATIONS = {
+        settings.LLM_SUMMARIZATION_MODEL_HRID: LLModel(
+            hrid="mistral-model",
+            model_name="mistral-7b-instruct-v0.1",
+            human_readable_name="Mistral 7B Instruct",
+            profile=None,
+            provider=LLMProvider(
+                hrid="mistral",
+                kind="mistral",
+                base_url="https://api.mistral.ai/v1",
+                api_key="testkey",
+            ),
+            is_active=True,
+            system_prompt="direct",
+            tools=[],
+        ),
+    }
+
+
+@pytest.fixture(name="mocked_context")
+def fixture_mocked_context():
+    """Fixture for a mocked RunContext."""
+    mock_ctx = mock.Mock(spec=RunContext)
+    mock_ctx.usage = RunUsage(input_tokens=0, output_tokens=0)
+    mock_ctx.max_retries = 2
+    mock_ctx.retries = {}
+    return mock_ctx
+
+
+def mocked_summary(_messages, _info=None):
+    """Mocked summary response."""
+    return ModelResponse(parts=[TextPart(content="This is a summary of the test chunk.")])
+
+
+@pytest.mark.asyncio
+async def test_summarize_chunk_returns_summary(mocked_context):
+    """Test that summarize_chunk returns a summary."""
+    summarization_agent = SummarizationAgent()
+
+    with summarization_agent.override(model=FunctionModel(mocked_summary)):
+        chunk = "This is a test chunk of text that needs to be summarized."
+
+        result = await summarize_chunk(1, chunk, 1, summarization_agent, mocked_context)
+
+        assert result == "This is a summary of the test chunk."
+
+
+@pytest.mark.asyncio
+async def test_summarize_chunk_raises_model_retry_on_error(mocked_context):
+    """Test that summarize_chunk raises ModelRetry when agent fails."""
+    summarization_agent = SummarizationAgent()
+
+    def mocked_summary_error(_messages, _info=None):
+        """Mocked summary that raises an error."""
+        raise ValueError("Simulated LLM error")
+
+    with summarization_agent.override(model=FunctionModel(mocked_summary_error)):
+        chunk = "This is a test chunk."
+
+        with pytest.raises(ModelRetry) as exc_info:
+            await summarize_chunk(1, chunk, 1, summarization_agent, mocked_context)
+
+        assert "An error occurred while summarizing a part of the document chunk" in str(
+            exc_info.value
+        )
+
+
+@pytest.mark.asyncio
+async def test_summarize_chunk_handles_empty_response(mocked_context):
+    """Test that summarize_chunk handles empty responses from the agent."""
+    summarization_agent = SummarizationAgent()
+
+    def mocked_empty_summary(_messages, _info=None):
+        """Mocked summary that returns empty content."""
+        return ModelResponse(parts=[TextPart(content="")])
+
+    with summarization_agent.override(model=FunctionModel(mocked_empty_summary)):
+        chunk = "This is a test chunk."
+
+        # Empty responses cause ModelRetry since pydantic-ai considers them invalid
+        with pytest.raises(ModelRetry):
+            await summarize_chunk(1, chunk, 1, summarization_agent, mocked_context)
+
+
+@pytest.mark.asyncio
+async def test_document_summarize_single_document(
+    settings, mocked_context, mock_summarization_agent
+):
+    """Test document_summarize with a single document."""
+    settings.SUMMARIZATION_CHUNK_SIZE = 100
+    settings.SUMMARIZATION_OVERLAP_SIZE = 10
+    settings.SUMMARIZATION_CONCURRENT_REQUESTS = 2
+
+    # Create mock conversation with a text attachment
+    mock_conversation = mock.Mock()
+    mock_attachment = mock.Mock()
+    mock_attachment.key = "test_doc.txt"
+    mock_attachment.file_name = "test_doc.txt"
+    mock_attachment.content_type = "text/plain"
+
+    mock_conversation.attachments.filter.return_value = [mock_attachment]
+
+    # Mock file storage
+    file_content = "This is a test document. " * 20  # Create a document with some content
+    with mock.patch.object(
+        default_storage, "open", return_value=io.BytesIO(file_content.encode("utf-8"))
+    ):
+        # Set up mocked_context with conversation
+        mocked_context.deps = mock.Mock()
+        mocked_context.deps.conversation = mock_conversation
+
+        call_count = {"chunk": 0, "merge": 0}
+
+        def mocked_summary_full(messages, _info=None):
+            """Mocked summary for full flow."""
+            messages_text = messages[0].parts[-1].content
+
+            if "Produce a coherent synthesis" in messages_text:
+                call_count["merge"] += 1
+                return ModelResponse(
+                    parts=[TextPart(content="# Final Summary\n\nThis is the final merged summary.")]
+                )
+
+            call_count["chunk"] += 1
+            return ModelResponse(
+                parts=[TextPart(content=f"Summary of chunk {call_count['chunk']}")]
+            )
+
+        with mock_summarization_agent(FunctionModel(mocked_summary_full)):
+            result = await document_summarize(mocked_context, instructions=None)
+
+        assert result.return_value == "# Final Summary\n\nThis is the final merged summary."
+        assert result.metadata["sources"] == {"test_doc.txt"}
+        assert call_count["merge"] == 1
+
+
+@pytest.mark.asyncio
+async def test_document_summarize_multiple_documents(
+    settings, mocked_context, mock_summarization_agent
+):
+    """Test document_summarize with multiple documents."""
+    settings.SUMMARIZATION_CHUNK_SIZE = 50
+    settings.SUMMARIZATION_OVERLAP_SIZE = 5
+    settings.SUMMARIZATION_CONCURRENT_REQUESTS = 2
+
+    # Create mock conversation with multiple text attachments
+    mock_conversation = mock.Mock()
+    mock_attachment1 = mock.Mock()
+    mock_attachment1.key = "doc1.txt"
+    mock_attachment1.file_name = "doc1.txt"
+    mock_attachment1.content_type = "text/plain"
+
+    mock_attachment2 = mock.Mock()
+    mock_attachment2.key = "doc2.txt"
+    mock_attachment2.file_name = "doc2.txt"
+    mock_attachment2.content_type = "text/plain"
+
+    mock_conversation.attachments.filter.return_value = [mock_attachment1, mock_attachment2]
+
+    file_content1 = "Content of document one. " * 10
+    file_content2 = "Content of document two. " * 10
+
+    def mock_open_side_effect(key):
+        """Mock file opening based on key."""
+        if key == "doc1.txt":
+            return io.BytesIO(file_content1.encode("utf-8"))
+        return io.BytesIO(file_content2.encode("utf-8"))
+
+    with mock.patch.object(default_storage, "open", side_effect=mock_open_side_effect):
+        # Set up mocked_context with conversation
+        mocked_context.deps = mock.Mock()
+        mocked_context.deps.conversation = mock_conversation
+
+        def mocked_summary_multi(messages, _info=None):
+            """Mocked summary for multiple documents."""
+            messages_text = messages[0].parts[-1].content
+
+            if "Produce a coherent synthesis" in messages_text:
+                return ModelResponse(parts=[TextPart(content="Combined summary of all documents")])
+
+            return ModelResponse(parts=[TextPart(content="Chunk summary")])
+
+        with mock_summarization_agent(FunctionModel(mocked_summary_multi)):
+            result = await document_summarize(mocked_context, instructions=None)
+
+        assert result.return_value == "Combined summary of all documents"
+        assert result.metadata["sources"] == {"doc1.txt", "doc2.txt"}
+
+
+@pytest.mark.asyncio
+async def test_document_summarize_with_custom_instructions(
+    settings, mocked_context, mock_summarization_agent
+):
+    """Test document_summarize with custom instructions."""
+    settings.SUMMARIZATION_CHUNK_SIZE = 100
+    settings.SUMMARIZATION_OVERLAP_SIZE = 10
+    settings.SUMMARIZATION_CONCURRENT_REQUESTS = 2
+
+    mock_conversation = mock.Mock()
+    mock_attachment = mock.Mock()
+    mock_attachment.key = "test.txt"
+    mock_attachment.file_name = "test.txt"
+    mock_attachment.content_type = "text/plain"
+
+    mock_conversation.attachments.filter.return_value = [mock_attachment]
+
+    file_content = "Test content " * 20
+
+    with mock.patch.object(
+        default_storage, "open", return_value=io.BytesIO(file_content.encode("utf-8"))
+    ):
+        # Set up mocked_context with conversation
+        mocked_context.deps = mock.Mock()
+        mocked_context.deps.conversation = mock_conversation
+
+        captured_merge_prompt = []
+
+        def mocked_summary_with_instructions(messages, _info=None):
+            """Mocked summary that captures merge prompt."""
+            messages_text = messages[0].parts[-1].content
+
+            if "Produce a coherent synthesis" in messages_text:
+                captured_merge_prompt.append(messages_text)
+                return ModelResponse(parts=[TextPart(content="Summary in 2 paragraphs")])
+
+            return ModelResponse(parts=[TextPart(content="Chunk summary")])
+
+        with mock_summarization_agent(FunctionModel(mocked_summary_with_instructions)):
+            result = await document_summarize(
+                mocked_context, instructions="summary in 2 paragraphs"
+            )
+
+        assert result.return_value == "Summary in 2 paragraphs"
+        assert len(captured_merge_prompt) == 1
+        assert "summary in 2 paragraphs" in captured_merge_prompt[0]
+
+
+@pytest.mark.asyncio
+async def test_document_summarize_no_text_attachments(mocked_context, mock_summarization_agent):
+    """Test document_summarize returns error message when no text documents found."""
+    mock_conversation = mock.Mock()
+    mock_conversation.attachments.filter.return_value = []
+
+    # Set up mocked_context with conversation
+    mocked_context.deps = mock.Mock()
+    mocked_context.deps.conversation = mock_conversation
+
+    # The decorator @last_model_retry_soft_fail catches ModelCannotRetry and returns a message
+    # We still need to provide a mock agent even if it won't be called
+    with mock_summarization_agent(FunctionModel(mocked_summary)):
+        result = await document_summarize(mocked_context, instructions=None)
+
+    assert "No text documents found in the conversation" in result
+
+
+@pytest.mark.asyncio
+async def test_document_summarize_error_reading_document(mocked_context, mock_summarization_agent):
+    """Test document_summarize handles errors when reading documents."""
+    mock_conversation = mock.Mock()
+    mock_attachment = mock.Mock()
+    mock_attachment.key = "test.txt"
+    mock_attachment.file_name = "test.txt"
+    mock_attachment.content_type = "text/plain"
+
+    mock_conversation.attachments.filter.return_value = [mock_attachment]
+
+    with mock.patch.object(default_storage, "open", side_effect=IOError("File read error")):
+        # Set up mocked_context with conversation
+        mocked_context.deps = mock.Mock()
+        mocked_context.deps.conversation = mock_conversation
+
+        # The decorator @last_model_retry_soft_fail catches ModelCannotRetry and returns a message
+        # We still need to provide a mock agent even if it won't be called
+        with mock_summarization_agent(FunctionModel(mocked_summary)):
+            result = await document_summarize(mocked_context, instructions=None)
+
+        assert "An unexpected error occurred during document summarization" in result
+
+
+@pytest.mark.asyncio
+async def test_document_summarize_error_during_chunk_summarization(
+    settings, mocked_context, mock_summarization_agent
+):
+    """Test document_summarize handles ModelRetry during chunk summarization."""
+    settings.SUMMARIZATION_CHUNK_SIZE = 100
+    settings.SUMMARIZATION_OVERLAP_SIZE = 10
+    settings.SUMMARIZATION_CONCURRENT_REQUESTS = 2
+
+    mock_conversation = mock.Mock()
+    mock_attachment = mock.Mock()
+    mock_attachment.key = "test.txt"
+    mock_attachment.file_name = "test.txt"
+    mock_attachment.content_type = "text/plain"
+
+    mock_conversation.attachments.filter.return_value = [mock_attachment]
+
+    file_content = "Test content " * 20
+
+    with mock.patch.object(
+        default_storage, "open", return_value=io.BytesIO(file_content.encode("utf-8"))
+    ):
+        # Set up mocked_context with conversation
+        mocked_context.deps = mock.Mock()
+        mocked_context.deps.conversation = mock_conversation
+
+        def mocked_summary_error(messages, _info=None):
+            """Mocked summary that raises an error during chunks."""
+            messages_text = messages[0].parts[-1].content
+
+            if "Produce a coherent synthesis" not in messages_text:
+                raise ValueError("Chunk processing error")
+
+            return ModelResponse(parts=[TextPart(content="Final summary")])
+
+        with mock_summarization_agent(FunctionModel(mocked_summary_error)):
+            with pytest.raises(ModelRetry):
+                await document_summarize(mocked_context, instructions=None)
+
+
+@pytest.mark.asyncio
+async def test_document_summarize_error_during_merge(
+    settings, mocked_context, mock_summarization_agent
+):
+    """Test document_summarize handles errors during final merge."""
+    settings.SUMMARIZATION_CHUNK_SIZE = 100
+    settings.SUMMARIZATION_OVERLAP_SIZE = 10
+    settings.SUMMARIZATION_CONCURRENT_REQUESTS = 2
+
+    mock_conversation = mock.Mock()
+    mock_attachment = mock.Mock()
+    mock_attachment.key = "test.txt"
+    mock_attachment.file_name = "test.txt"
+    mock_attachment.content_type = "text/plain"
+
+    mock_conversation.attachments.filter.return_value = [mock_attachment]
+
+    file_content = "Test content " * 20
+
+    with mock.patch.object(
+        default_storage, "open", return_value=io.BytesIO(file_content.encode("utf-8"))
+    ):
+        # Set up mocked_context with conversation
+        mocked_context.deps = mock.Mock()
+        mocked_context.deps.conversation = mock_conversation
+
+        def mocked_summary_merge_error(messages, _info=None):
+            """Mocked summary that raises an error during merge."""
+            messages_text = messages[0].parts[-1].content
+
+            if "Produce a coherent synthesis" in messages_text:
+                raise ValueError("Merge error")
+
+            return ModelResponse(parts=[TextPart(content="Chunk summary")])
+
+        with mock_summarization_agent(FunctionModel(mocked_summary_merge_error)):
+            with pytest.raises(ModelRetry) as exc_info:
+                await document_summarize(mocked_context, instructions=None)
+
+            # Should raise ModelRetry regardless of which phase failed
+            assert "An error occurred" in str(exc_info.value)
+
+
+@pytest.mark.asyncio
+async def test_document_summarize_empty_result(settings, mocked_context, mock_summarization_agent):
+    """Test document_summarize raises ModelRetry when summarization produces empty result."""
+    settings.SUMMARIZATION_CHUNK_SIZE = 100
+    settings.SUMMARIZATION_OVERLAP_SIZE = 10
+    settings.SUMMARIZATION_CONCURRENT_REQUESTS = 2
+
+    mock_conversation = mock.Mock()
+    mock_attachment = mock.Mock()
+    mock_attachment.key = "test.txt"
+    mock_attachment.file_name = "test.txt"
+    mock_attachment.content_type = "text/plain"
+
+    mock_conversation.attachments.filter.return_value = [mock_attachment]
+
+    file_content = "Test content " * 20
+
+    with mock.patch.object(
+        default_storage, "open", return_value=io.BytesIO(file_content.encode("utf-8"))
+    ):
+        # Set up mocked_context with conversation
+        mocked_context.deps = mock.Mock()
+        mocked_context.deps.conversation = mock_conversation
+
+        def mocked_empty_summary(messages, _info=None):
+            """Mocked summary that returns empty for merge."""
+            messages_text = messages[0].parts[-1].content
+
+            if "Produce a coherent synthesis" in messages_text:
+                return ModelResponse(parts=[TextPart(content="   ")])
+
+            return ModelResponse(parts=[TextPart(content="Chunk summary")])
+
+        with mock_summarization_agent(FunctionModel(mocked_empty_summary)):
+            with pytest.raises(ModelRetry) as exc_info:
+                await document_summarize(mocked_context, instructions=None)
+
+            # Should raise ModelRetry with the specific message
+            assert "The summarization produced an empty result" in str(exc_info.value)
+
+
+@pytest.mark.asyncio
+async def test_document_summarize_large_document_multiple_chunks(
+    settings, mocked_context, mock_summarization_agent
+):
+    """Test document_summarize with a large document requiring multiple chunks."""
+    settings.SUMMARIZATION_CHUNK_SIZE = 20  # Small chunk size to force multiple chunks
+    settings.SUMMARIZATION_OVERLAP_SIZE = 5
+    settings.SUMMARIZATION_CONCURRENT_REQUESTS = 2
+
+    mock_conversation = mock.Mock()
+    mock_attachment = mock.Mock()
+    mock_attachment.key = "large_doc.txt"
+    mock_attachment.file_name = "large_doc.txt"
+    mock_attachment.content_type = "text/plain"
+
+    mock_conversation.attachments.filter.return_value = [mock_attachment]
+
+    # Create a large document
+    file_content = "This is a word. " * 100  # Should create multiple chunks
+
+    with mock.patch.object(
+        default_storage, "open", return_value=io.BytesIO(file_content.encode("utf-8"))
+    ):
+        # Set up mocked_context with conversation
+        mocked_context.deps = mock.Mock()
+        mocked_context.deps.conversation = mock_conversation
+
+        chunk_count = {"count": 0}
+
+        def mocked_summary_multi_chunks(messages, _info=None):
+            """Mocked summary that counts chunks."""
+            messages_text = messages[0].parts[-1].content
+
+            if "Produce a coherent synthesis" in messages_text:
+                return ModelResponse(
+                    parts=[TextPart(content=f"Final summary of {chunk_count['count']} chunks")]
+                )
+
+            chunk_count["count"] += 1
+            return ModelResponse(
+                parts=[TextPart(content=f"Summary of chunk {chunk_count['count']}")]
+            )
+
+        with mock_summarization_agent(FunctionModel(mocked_summary_multi_chunks)):
+            result = await document_summarize(mocked_context, instructions=None)
+
+        assert "Final summary of" in result.return_value
+        assert chunk_count["count"] > 1  # Ensure multiple chunks were processed

src/backend/chat/tests/views/chat/conversations/test_conversation_with_document_upload.py

@@ -151,7 +151,7 @@ def fixture_mock_summarization_agent():
             super().__init__(**kwargs)
             self._model = FunctionModel(function=summarization_model)  # pylint: disable=protected-access
 
-    with mock.patch("chat.agents.summarize.SummarizationAgent", new=SummarizationAgentMock):
+    with mock.patch("chat.tools.document_summarize.SummarizationAgent", new=SummarizationAgentMock):
         yield
 
 
@@ -374,11 +374,10 @@ def test_post_conversation_with_document_upload(  # pylint: disable=too-many-arg
                 "timestamp": timezone_now,
             },
             {
-                "content": "If the user wants specific information from a "
-                "document, invoke web_search_albert_rag with an "
-                "appropriate query string.Do not ask the user for the "
-                "document; rely on the tool to locate and return "
-                "relevant passages.",
+                "content": "Use document_search_rag ONLY to retrieve specific "
+                "passages from attached documents. Do NOT use it to "
+                "summarize; for summaries, call the summarize tool "
+                "instead.",
                 "dynamic_ref": None,
                 "part_kind": "system-prompt",
                 "timestamp": timezone_now,
@@ -397,6 +396,15 @@ def test_post_conversation_with_document_upload(  # pylint: disable=too-many-arg
                 "part_kind": "system-prompt",
                 "timestamp": timezone_now,
             },
+            {
+                "content": "[Internal context] User documents are attached to this "
+                "conversation. Do not request re-upload of documents; "
+                "consider them already available via the internal "
+                "store.",
+                "dynamic_ref": None,
+                "part_kind": "system-prompt",
+                "timestamp": timezone_now,
+            },
             {
                 "content": ["What does the document say?"],
                 "part_kind": "user-prompt",
@@ -627,7 +635,7 @@ def test_post_conversation_with_document_upload_summarize(  # pylint: disable=to
         'document discusses various topics."}\n'
         '0:"The document discusses various topics."\n'
         'f:{"messageId":"<mocked_uuid>"}\n'
-        'd:{"finishReason":"stop","usage":{"promptTokens":201,"completionTokens":13}}\n'
+        'd:{"finishReason":"stop","usage":{"promptTokens":317,"completionTokens":19}}\n'
     )
 
     # Check that the conversation was updated
@@ -709,11 +717,10 @@ def test_post_conversation_with_document_upload_summarize(  # pylint: disable=to
                 "timestamp": timezone_now,
             },
             {
-                "content": "If the user wants specific information from a "
-                "document, invoke web_search_albert_rag with an "
-                "appropriate query string.Do not ask the user for the "
-                "document; rely on the tool to locate and return "
-                "relevant passages.",
+                "content": "Use document_search_rag ONLY to retrieve specific "
+                "passages from attached documents. Do NOT use it to "
+                "summarize; for summaries, call the summarize tool "
+                "instead.",
                 "dynamic_ref": None,
                 "part_kind": "system-prompt",
                 "timestamp": timezone_now,
@@ -732,6 +739,15 @@ def test_post_conversation_with_document_upload_summarize(  # pylint: disable=to
                 "part_kind": "system-prompt",
                 "timestamp": timezone_now,
             },
+            {
+                "content": "[Internal context] User documents are attached to this "
+                "conversation. Do not request re-upload of documents; "
+                "consider them already available via the internal "
+                "store.",
+                "dynamic_ref": None,
+                "part_kind": "system-prompt",
+                "timestamp": timezone_now,
+            },
             {
                 "content": ["Make a summary of this document."],
                 "part_kind": "user-prompt",

src/backend/chat/tests/views/chat/conversations/test_conversation_with_document_url.py

@@ -881,10 +881,9 @@ def test_post_conversation_with_local_not_pdf_document_url(  # pylint: disable=t
                     SystemPromptPart(content="Answer in english.", timestamp=timezone.now()),
                     SystemPromptPart(
                         content=(
-                            "If the user wants specific information from a document, "
-                            "invoke web_search_albert_rag with an appropriate query string."
-                            "Do not ask the user for the document; rely on the tool to locate "
-                            "and return relevant passages."
+                            "Use document_search_rag ONLY to retrieve specific passages from "
+                            "attached documents. Do NOT use it to summarize; for summaries, "
+                            "call the summarize tool instead."
                         ),
                         timestamp=timezone.now(),
                     ),
@@ -901,6 +900,14 @@ def test_post_conversation_with_local_not_pdf_document_url(  # pylint: disable=t
                         ),
                         timestamp=timezone.now(),
                     ),
+                    SystemPromptPart(
+                        content=(
+                            "[Internal context] User documents are attached to this conversation. "
+                            "Do not request re-upload of documents; consider them already "
+                            "available via the internal store."
+                        ),
+                        timestamp=timezone.now(),
+                    ),
                     UserPromptPart(
                         content=[
                             "What is in this document?",
@@ -1002,11 +1009,10 @@ def test_post_conversation_with_local_not_pdf_document_url(  # pylint: disable=t
                     "timestamp": timestamp,
                 },
                 {
-                    "content": "If the user wants specific information from a "
-                    "document, invoke web_search_albert_rag with an "
-                    "appropriate query string.Do not ask the user for the "
-                    "document; rely on the tool to locate and return "
-                    "relevant passages.",
+                    "content": "Use document_search_rag ONLY to retrieve specific "
+                    "passages from attached documents. Do NOT use it to "
+                    "summarize; for summaries, call the summarize tool "
+                    "instead.",
                     "dynamic_ref": None,
                     "part_kind": "system-prompt",
                     "timestamp": timestamp,
@@ -1025,6 +1031,15 @@ def test_post_conversation_with_local_not_pdf_document_url(  # pylint: disable=t
                     "part_kind": "system-prompt",
                     "timestamp": timestamp,
                 },
+                {
+                    "content": "[Internal context] User documents are attached to "
+                    "this conversation. Do not request re-upload of "
+                    "documents; consider them already available via the "
+                    "internal store.",
+                    "dynamic_ref": None,
+                    "part_kind": "system-prompt",
+                    "timestamp": timestamp,
+                },
                 {
                     "content": [
                         "What is in this document?",

src/backend/chat/tests/views/chat/test_media_auth.py

@@ -25,6 +25,8 @@ def test_api_media_auth_unkown_document(api_client):
     Trying to download a media related to a conversation that does not exist
     should not have the side effect to create it (no regression test).
     """
+    ChatConversation.objects.all().delete()
+
     original_url = f"http://localhost/media/{uuid4()!s}/attachments/{uuid4()!s}.jpg"
 
     response = api_client.get("/api/v1.0/chats/media-auth/", HTTP_X_ORIGINAL_URL=original_url)

src/backend/chat/tools/document_search_rag.py

@@ -20,7 +20,7 @@ def add_document_rag_search_tool(agent: Agent) -> None:
 
         Args:
             ctx (RunContext): The run context containing the conversation.
-            query (str): The term to search the internet for.
+            query (str): The query to search the documents for.
         """
         document_store_backend = import_string(settings.RAG_DOCUMENT_SEARCH_BACKEND)
 
@@ -43,8 +43,6 @@ def add_document_rag_search_tool(agent: Agent) -> None:
     def document_rag_instructions() -> str:
         """Dynamic system prompt function to add RAG instructions if any."""
         return (
-            "If the user wants specific information from a document, invoke "
-            "web_search_albert_rag with an appropriate query string."
-            "Do not ask the user for the document; rely on the tool to locate "
-            "and return relevant passages."
+            "Use document_search_rag ONLY to retrieve specific passages from attached documents. "
+            "Do NOT use it to summarize; for summaries, call the summarize tool instead."
         )

src/backend/chat/tools/document_summarize.py

@@ -0,0 +1,189 @@
+"""Summarization tool used for uploaded documents."""
+
+import asyncio
+import logging
+
+from django.conf import settings
+from django.core.files.storage import default_storage
+
+import semchunk
+from asgiref.sync import sync_to_async
+from pydantic_ai import RunContext
+from pydantic_ai.exceptions import ModelRetry
+from pydantic_ai.messages import ToolReturn
+
+from chat.agents.summarize import SummarizationAgent
+from chat.tools.exceptions import ModelCannotRetry
+from chat.tools.utils import last_model_retry_soft_fail
+
+logger = logging.getLogger(__name__)
+
+
+@sync_to_async
+def read_document_content(doc):
+    """Read document content asynchronously."""
+    with default_storage.open(doc.key) as f:
+        return doc.file_name, f.read().decode("utf-8")
+
+
+async def summarize_chunk(idx, chunk, total_chunks, summarization_agent, ctx):
+    """Summarize a single chunk of text."""
+    sum_prompt = (
+        "You are an agent specializing in text summarization. "
+        "Generate a clear and concise summary of the following passage "
+        f"(part {idx}/{total_chunks}):\n'''\n{chunk}\n'''\n\n"
+    )
+
+    logger.debug(
+        "[summarize] CHUNK %s/%s prompt=> %s", idx, total_chunks, sum_prompt[0:100] + "..."
+    )
+
+    try:
+        resp = await summarization_agent.run(sum_prompt, usage=ctx.usage)
+    except Exception as exc:
+        logger.warning("Error during chunk summarization: %s", exc, exc_info=True)
+        raise ModelRetry(
+            "An error occurred while summarizing a part of the document chunk."
+        ) from exc
+
+    logger.debug("[summarize] CHUNK %s/%s response<= %s", idx, total_chunks, resp.output or "")
+    return resp.output or ""
+
+
+@last_model_retry_soft_fail
+async def document_summarize(  # pylint: disable=too-many-locals
+    ctx: RunContext, *, instructions: str | None = None
+) -> ToolReturn:
+    """
+    Generate a complete, ready-to-use summary of the documents in context
+    (do not request the documents to the user).
+    Return this summary directly to the user WITHOUT any modification,
+    or additional summarization.
+    The summary is already optimized and MUST be presented as-is in the final response
+    or translated preserving the information.
+
+    Instructions are optional but should reflect the user's request.
+
+    Examples:
+    "Summarize this doc in 2 paragraphs" -> instructions = "summary in 2 paragraphs"
+    "Summarize this doc in English" -> instructions = "In English"
+    "Summarize this doc" -> instructions = "" (default)
+
+    Args:
+        instructions (str | None): The instructions the user gave to use for the summarization
+    """
+    try:
+        instructions_hint = (
+            instructions.strip() if instructions else "The summary should contain 2 or 3 parts."
+        )
+        summarization_agent = SummarizationAgent()
+
+        # Collect documents content
+        text_attachment = await sync_to_async(list)(
+            ctx.deps.conversation.attachments.filter(
+                content_type__startswith="text/",
+            )
+        )
+
+        if not text_attachment:
+            raise ModelCannotRetry(
+                "No text documents found in the conversation. "
+                "You must explain this to the user and ask them to provide documents."
+            )
+
+        documents = [await read_document_content(doc) for doc in text_attachment]
+
+        # Chunk documents and summarize each chunk
+        chunk_size = settings.SUMMARIZATION_CHUNK_SIZE
+        chunker = semchunk.chunkerify(
+            tokenizer_or_token_counter=lambda text: len(text.split()),
+            chunk_size=chunk_size,
+        )
+        documents_chunks = chunker(
+            [doc[1] for doc in documents],
+            overlap=settings.SUMMARIZATION_OVERLAP_SIZE,
+        )
+
+        logger.info(
+            "[summarize] chunking: %s parts (size~%s), instructions='%s'",
+            sum(len(chunks) for chunks in documents_chunks),
+            chunk_size,
+            instructions_hint,
+        )
+
+        # Parallelize the chunk summarization with a semaphore to limit concurrent tasks
+        # because it can be very resource intensive on the LLM backend
+        semaphore = asyncio.Semaphore(settings.SUMMARIZATION_CONCURRENT_REQUESTS)
+
+        async def summarize_chunk_with_semaphore(idx, chunk, total_chunks):
+            """Summarize a chunk with semaphore-controlled concurrency."""
+            async with semaphore:
+                return await summarize_chunk(idx, chunk, total_chunks, summarization_agent, ctx)
+
+        doc_chunk_summaries = []
+        try:
+            for doc_chunks in documents_chunks:
+                summarization_tasks = [
+                    summarize_chunk_with_semaphore(idx, chunk, len(doc_chunks))
+                    for idx, chunk in enumerate(doc_chunks, start=1)
+                ]
+                chunk_summaries = await asyncio.gather(*summarization_tasks)
+                doc_chunk_summaries.append(chunk_summaries)
+        except ModelRetry as exc:
+            logger.warning("Retryable error during chunk summarization: %s", exc, exc_info=True)
+            raise
+        except Exception as exc:
+            logger.warning("Error during chunk summarization: %s", exc, exc_info=True)
+            raise ModelRetry("An error occurred while processing document chunks.") from exc
+
+        context = "\n\n".join(
+            doc_name + "\n\n" + "\n\n".join(summaries)
+            for doc_name, summaries in zip(
+                (doc[0] for doc in documents),
+                doc_chunk_summaries,
+                strict=True,
+            )
+        )
+
+        # Merge chunk summaries into a single concise summary
+        merged_prompt = (
+            "Produce a coherent synthesis from the summaries below.\n\n"
+            f"'''\n{context}\n'''\n\n"
+            "Constraints:\n"
+            "- Summarize without repetition.\n"
+            "- Harmonize style and terminology.\n"
+            "- The final summary must be well-structured and formatted in markdown.\n"
+            f"- Follow the instructions: {instructions_hint}\n"
+            "Respond directly with the final summary."
+        )
+
+        logger.debug("[summarize] MERGE prompt=> %s", merged_prompt)
+
+        try:
+            merged_resp = await summarization_agent.run(merged_prompt, usage=ctx.usage)
+        except Exception as exc:
+            logger.warning("Error during merge summarization: %s", exc, exc_info=True)
+            raise ModelRetry("An error occurred while generating the final summary.") from exc
+
+        final_summary = (merged_resp.output or "").strip()
+
+        if not final_summary:
+            raise ModelRetry("The summarization produced an empty result.")
+
+        logger.debug("[summarize] MERGE response<= %s", final_summary)
+
+        return ToolReturn(
+            return_value=final_summary,
+            metadata={"sources": {doc[0] for doc in documents}},
+        )
+
+    except (ModelCannotRetry, ModelRetry):
+        # Re-raise these as-is
+        raise
+    except Exception as exc:
+        # Unexpected error - stop and inform user
+        logger.exception("Unexpected error in document_summarize: %s", exc)
+        raise ModelCannotRetry(
+            f"An unexpected error occurred during document summarization: {type(exc).__name__}. "
+            "You must explain this to the user and not try to answer based on your knowledge."
+        ) from exc

src/backend/conversations/settings.py

@@ -312,7 +312,7 @@ class Base(BraveSettings, Configuration):
         "django.middleware.common.CommonMiddleware",
         "django.middleware.csrf.CsrfViewMiddleware",
         "django.contrib.auth.middleware.AuthenticationMiddleware",
-        "core.posthog.AsyncPosthogContextMiddleware",
+        "posthog.integrations.django.PosthogContextMiddleware",
         "django.contrib.messages.middleware.MessageMiddleware",
         "dockerflow.django.middleware.DockerflowMiddleware",
     ]
@@ -786,6 +786,21 @@ USER QUESTION:
         environ_name="SUMMARIZATION_SYSTEM_PROMPT",
         environ_prefix=None,
     )
+    SUMMARIZATION_CHUNK_SIZE = values.PositiveIntegerValue(
+        default=20_000,  # Approx 20k words per chunk
+        environ_name="SUMMARIZATION_CHUNK_SIZE",
+        environ_prefix=None,
+    )
+    SUMMARIZATION_OVERLAP_SIZE = values.FloatValue(
+        default=0.05,  # 5% overlap
+        environ_name="SUMMARIZATION_OVERLAP_SIZE",
+        environ_prefix=None,
+    )
+    SUMMARIZATION_CONCURRENT_REQUESTS = values.PositiveIntegerValue(
+        default=5,
+        environ_name="SUMMARIZATION_CONCURRENT_REQUESTS",
+        environ_prefix=None,
+    )
 
     # Tavily API
     TAVILY_API_KEY = values.Value(

src/backend/core/posthog.py

@@ -1,113 +0,0 @@
-"""
-Asynchronous PostHog context middleware for Django.
-
-Since https://github.com/PostHog/posthog-python/commit/6af129f41413f4f7d55731763ff42e4c0fb66844
-The official PostHog Django middleware supports both sync and async requests,
-but the call to request.user fails in async contexts.
-
-This is an horrible patch while waiting for an official fix.
-Follow https://github.com/PostHog/posthog-python/issues/355
-"""
-
-from posthog import contexts
-from posthog.integrations.django import PosthogContextMiddleware
-
-
-class AsyncPosthogContextMiddleware(PosthogContextMiddleware):
-    """
-    Asynchronous Django middleware to extract PostHog context from HTTP requests.
-
-    While the original PosthogContextMiddleware is supposed to manage both sync and async requests,
-    the call to request.user fails in async contexts.
-    """
-
-    async def extract_tags_async(self, request):
-        """Extract tags from the HTTP request asynchronously."""
-        tags = {}
-
-        (user_id, user_email) = await self.extract_request_user_async(request)
-
-        # Extract session ID from X-POSTHOG-SESSION-ID header
-        session_id = request.headers.get("X-POSTHOG-SESSION-ID")
-        if session_id:
-            contexts.set_context_session(session_id)
-
-        # Extract distinct ID from X-POSTHOG-DISTINCT-ID header or request user id
-        distinct_id = request.headers.get("X-POSTHOG-DISTINCT-ID") or user_id
-        if distinct_id:
-            contexts.identify_context(distinct_id)
-
-        # Extract user email
-        if user_email:
-            tags["email"] = user_email
-
-        # Extract current URL
-        absolute_url = request.build_absolute_uri()
-        if absolute_url:
-            tags["$current_url"] = absolute_url
-
-        # Extract request method
-        if request.method:
-            tags["$request_method"] = request.method
-
-        # Extract request path
-        if request.path:
-            tags["$request_path"] = request.path
-
-        # Extract IP address
-        ip_address = request.headers.get("X-Forwarded-For")
-        if ip_address:
-            tags["$ip_address"] = ip_address
-
-        # Extract user agent
-        user_agent = request.headers.get("User-Agent")
-        if user_agent:
-            tags["$user_agent"] = user_agent
-
-        # Apply extra tags if configured
-        if self.extra_tags:
-            extra = self.extra_tags(request)
-            if extra:
-                tags.update(extra)
-
-        # Apply tag mapping if configured
-        if self.tag_map:
-            tags = self.tag_map(tags)
-
-        return tags
-
-    async def extract_request_user_async(self, request):
-        """Extract user ID and email from the HTTP request asynchronously."""
-        user_id = None
-        email = None
-
-        user = await request.auser()
-
-        if user and getattr(user, "is_authenticated", False):
-            try:
-                user_id = str(user.pk)
-            except Exception:  # noqa: BLE001, S110  # pylint: disable=broad-except
-                pass
-
-            try:
-                email = str(user.email)
-            except Exception:  # noqa: BLE001, S110  # pylint: disable=broad-except
-                pass
-
-        return user_id, email
-
-    async def __acall__(self, request):
-        """
-        Asynchronous entry point for async request handling.
-
-        This method is called when the middleware chain is async.
-        """
-        if self.request_filter and not self.request_filter(request):
-            return await self.get_response(request)
-
-        with contexts.new_context(self.capture_exceptions, client=self.client):
-            tags = await self.extract_tags_async(request)
-            for k, v in tags.items():
-                contexts.tag(k, v)
-
-            return await self.get_response(request)

src/backend/pyproject.toml

@@ -27,41 +27,42 @@ requires-python = ">=3.12"
 dependencies = [
     "deprecated",
     "beautifulsoup4==4.14.2",
-    "boto3==1.40.59",
-    "Brotli==1.1.0",
+    "boto3==1.40.67",
+    "Brotli==1.2.0",
     "django-configurations==2.5.1",
     "django-cors-headers==4.9.0",
-    "django-countries==7.6.1",
+    "django-countries==8.0.0",
     "django-filter==25.2",
-    "django-lasuite[all]==0.0.16",
+    "django-lasuite[all]==0.0.17",
     "django-parler==2.3",
     "django-pydantic-field==0.3.13",
     "django-redis==6.0.0",
     "django-storages[s3]==1.14.6",
     "django-timezone-field>=5.1",
-    "django==5.2.7",
+    "django==5.2.8",
     "djangorestframework==3.16.1",
-    "drf_spectacular==0.28.0",
+    "drf_spectacular==0.29.0",
     "dockerflow==2024.4.2",
     "easy_thumbnails==2.10.1",
     "factory_boy==3.3.3",
     "gunicorn==23.0.0",
     "jsonschema==4.25.1",
-    "langfuse==3.8.1",
+    "langfuse==3.9.0",
     "lxml==5.4.0",
-    "markdown==3.9",
+    "markdown==3.10",
     "markitdown==0.0.2",
     "mozilla-django-oidc==4.0.1",
     "nested-multipart-parser==1.6.0",
-    "posthog==6.7.10",
-    "pydantic==2.12.3",
-    "pydantic-ai-slim[openai,mistral,mcp,evals,logfire]==1.6.0",
+    "posthog==6.8.0",
+    "pydantic==2.12.4",
+    "pydantic-ai-slim[openai,mistral,mcp,evals,logfire]==1.11.0",
     "psycopg[binary]==3.2.12",
     "PyJWT==2.10.1",
     "python-magic==0.4.27",
     "redis<6.0.0",
     "requests==2.32.5",
-    "sentry-sdk==2.42.1",
+    "semchunk==3.2.5",
+    "sentry-sdk==2.43.0",
     "trafilatura==2.0.0",
     "uvicorn==0.38.0",
     "whitenoise==6.11.0",
@@ -81,11 +82,11 @@ dev = [
     "drf-spectacular-sidecar==2025.10.1",
     "freezegun==1.5.5",
     "ipdb==0.13.13",
-    "ipython==9.6.0",
-    "pyfakefs==5.10.0",
+    "ipython==9.7.0",
+    "pyfakefs==5.10.2",
     "pylint-django==2.6.1",
     "pylint==3.3.9",
-    "pylint-pydantic==0.4.0",
+    "pylint-pydantic==0.4.1",
     "pytest-asyncio==1.2.0",
     "pytest-cov==7.0.0",
     "pytest-django==4.11.1",
@@ -94,7 +95,7 @@ dev = [
     "pytest-xdist==3.8.0",
     "responses==0.25.8",
     "respx==0.22.0",
-    "ruff==0.14.2",
+    "ruff==0.14.3",
     "types-requests==2.32.4.20250913",
 ]
 

src/frontend/apps/conversations/src/features/chat/components/Chat.tsx

@@ -701,7 +701,23 @@ export const Chat = ({
                             >
                               <Loader />
                               <Text $variation="600" $size="md">
-                                {t('Search...')}
+                                {(() => {
+                                  const toolInvocation = message.parts?.find(
+                                    (part) =>
+                                      part.type === 'tool-invocation' &&
+                                      part.toolInvocation.toolName !==
+                                        'document_parsing',
+                                  );
+                                  if (
+                                    toolInvocation?.type ===
+                                      'tool-invocation' &&
+                                    toolInvocation.toolInvocation.toolName ===
+                                      'summarize'
+                                  ) {
+                                    return t('Summarizing...');
+                                  }
+                                  return t('Search...');
+                                })()}
                               </Text>
                             </Box>
                           )}