lite node

exo theme
EXO iOS app
2026-02-19 22:07:34 +05:00 · 2026-02-19 04:21:58 +05:00 · 2026-02-18 06:40:07 +05:00 · 2026-02-17 14:31:47 +00:00 · 2026-02-17 11:48:28 +00:00 · 2026-02-17 11:43:06 +00:00
57 changed files with 4282 additions and 753 deletions
@@ -0,0 +1,628 @@
+// !$*UTF8*$!
+{
+	archiveVersion = 1;
+	classes = {
+	};
+	objectVersion = 77;
+	objects = {
+
+/* Begin PBXBuildFile section */
+		E09D17522F44F359009C51A3 /* MLXLLM in Frameworks */ = {isa = PBXBuildFile; productRef = E09D17512F44F359009C51A3 /* MLXLLM */; };
+		E09D17542F44F359009C51A3 /* MLXLMCommon in Frameworks */ = {isa = PBXBuildFile; productRef = E09D17532F44F359009C51A3 /* MLXLMCommon */; };
+/* End PBXBuildFile section */
+
+/* Begin PBXContainerItemProxy section */
+		E09D167D2F44CA20009C51A3 /* PBXContainerItemProxy */ = {
+			isa = PBXContainerItemProxy;
+			containerPortal = E09D16672F44CA1E009C51A3 /* Project object */;
+			proxyType = 1;
+			remoteGlobalIDString = E09D166E2F44CA1E009C51A3;
+			remoteInfo = "EXO-iOS";
+		};
+		E09D16872F44CA20009C51A3 /* PBXContainerItemProxy */ = {
+			isa = PBXContainerItemProxy;
+			containerPortal = E09D16672F44CA1E009C51A3 /* Project object */;
+			proxyType = 1;
+			remoteGlobalIDString = E09D166E2F44CA1E009C51A3;
+			remoteInfo = "EXO-iOS";
+		};
+/* End PBXContainerItemProxy section */
+
+/* Begin PBXFileReference section */
+		E09D166F2F44CA1E009C51A3 /* EXO-iOS.app */ = {isa = PBXFileReference; explicitFileType = wrapper.application; includeInIndex = 0; path = "EXO-iOS.app"; sourceTree = BUILT_PRODUCTS_DIR; };
+		E09D167C2F44CA20009C51A3 /* EXO-iOSTests.xctest */ = {isa = PBXFileReference; explicitFileType = wrapper.cfbundle; includeInIndex = 0; path = "EXO-iOSTests.xctest"; sourceTree = BUILT_PRODUCTS_DIR; };
+		E09D16862F44CA20009C51A3 /* EXO-iOSUITests.xctest */ = {isa = PBXFileReference; explicitFileType = wrapper.cfbundle; includeInIndex = 0; path = "EXO-iOSUITests.xctest"; sourceTree = BUILT_PRODUCTS_DIR; };
+/* End PBXFileReference section */
+
+/* Begin PBXFileSystemSynchronizedBuildFileExceptionSet section */
+		E09D169A2F44CA20009C51A3 /* Exceptions for "EXO-iOS" folder in "EXO-iOS" target */ = {
+			isa = PBXFileSystemSynchronizedBuildFileExceptionSet;
+			membershipExceptions = (
+				Info.plist,
+			);
+			target = E09D166E2F44CA1E009C51A3 /* EXO-iOS */;
+		};
+/* End PBXFileSystemSynchronizedBuildFileExceptionSet section */
+
+/* Begin PBXFileSystemSynchronizedRootGroup section */
+		E09D16712F44CA1E009C51A3 /* EXO-iOS */ = {
+			isa = PBXFileSystemSynchronizedRootGroup;
+			exceptions = (
+				E09D169A2F44CA20009C51A3 /* Exceptions for "EXO-iOS" folder in "EXO-iOS" target */,
+			);
+			path = "EXO-iOS";
+			sourceTree = "<group>";
+		};
+		E09D167F2F44CA20009C51A3 /* EXO-iOSTests */ = {
+			isa = PBXFileSystemSynchronizedRootGroup;
+			path = "EXO-iOSTests";
+			sourceTree = "<group>";
+		};
+		E09D16892F44CA20009C51A3 /* EXO-iOSUITests */ = {
+			isa = PBXFileSystemSynchronizedRootGroup;
+			path = "EXO-iOSUITests";
+			sourceTree = "<group>";
+		};
+/* End PBXFileSystemSynchronizedRootGroup section */
+
+/* Begin PBXFrameworksBuildPhase section */
+		E09D166C2F44CA1E009C51A3 /* Frameworks */ = {
+			isa = PBXFrameworksBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+				E09D17542F44F359009C51A3 /* MLXLMCommon in Frameworks */,
+				E09D17522F44F359009C51A3 /* MLXLLM in Frameworks */,
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+		E09D16792F44CA20009C51A3 /* Frameworks */ = {
+			isa = PBXFrameworksBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+		E09D16832F44CA20009C51A3 /* Frameworks */ = {
+			isa = PBXFrameworksBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+/* End PBXFrameworksBuildPhase section */
+
+/* Begin PBXGroup section */
+		E09D16662F44CA1E009C51A3 = {
+			isa = PBXGroup;
+			children = (
+				E09D16712F44CA1E009C51A3 /* EXO-iOS */,
+				E09D167F2F44CA20009C51A3 /* EXO-iOSTests */,
+				E09D16892F44CA20009C51A3 /* EXO-iOSUITests */,
+				E09D16702F44CA1E009C51A3 /* Products */,
+			);
+			sourceTree = "<group>";
+		};
+		E09D16702F44CA1E009C51A3 /* Products */ = {
+			isa = PBXGroup;
+			children = (
+				E09D166F2F44CA1E009C51A3 /* EXO-iOS.app */,
+				E09D167C2F44CA20009C51A3 /* EXO-iOSTests.xctest */,
+				E09D16862F44CA20009C51A3 /* EXO-iOSUITests.xctest */,
+			);
+			name = Products;
+			sourceTree = "<group>";
+		};
+/* End PBXGroup section */
+
+/* Begin PBXNativeTarget section */
+		E09D166E2F44CA1E009C51A3 /* EXO-iOS */ = {
+			isa = PBXNativeTarget;
+			buildConfigurationList = E09D16902F44CA20009C51A3 /* Build configuration list for PBXNativeTarget "EXO-iOS" */;
+			buildPhases = (
+				E09D166B2F44CA1E009C51A3 /* Sources */,
+				E09D166C2F44CA1E009C51A3 /* Frameworks */,
+				E09D166D2F44CA1E009C51A3 /* Resources */,
+			);
+			buildRules = (
+			);
+			dependencies = (
+			);
+			fileSystemSynchronizedGroups = (
+				E09D16712F44CA1E009C51A3 /* EXO-iOS */,
+			);
+			name = "EXO-iOS";
+			packageProductDependencies = (
+				E09D17512F44F359009C51A3 /* MLXLLM */,
+				E09D17532F44F359009C51A3 /* MLXLMCommon */,
+			);
+			productName = "EXO-iOS";
+			productReference = E09D166F2F44CA1E009C51A3 /* EXO-iOS.app */;
+			productType = "com.apple.product-type.application";
+		};
+		E09D167B2F44CA20009C51A3 /* EXO-iOSTests */ = {
+			isa = PBXNativeTarget;
+			buildConfigurationList = E09D16932F44CA20009C51A3 /* Build configuration list for PBXNativeTarget "EXO-iOSTests" */;
+			buildPhases = (
+				E09D16782F44CA20009C51A3 /* Sources */,
+				E09D16792F44CA20009C51A3 /* Frameworks */,
+				E09D167A2F44CA20009C51A3 /* Resources */,
+			);
+			buildRules = (
+			);
+			dependencies = (
+				E09D167E2F44CA20009C51A3 /* PBXTargetDependency */,
+			);
+			fileSystemSynchronizedGroups = (
+				E09D167F2F44CA20009C51A3 /* EXO-iOSTests */,
+			);
+			name = "EXO-iOSTests";
+			packageProductDependencies = (
+			);
+			productName = "EXO-iOSTests";
+			productReference = E09D167C2F44CA20009C51A3 /* EXO-iOSTests.xctest */;
+			productType = "com.apple.product-type.bundle.unit-test";
+		};
+		E09D16852F44CA20009C51A3 /* EXO-iOSUITests */ = {
+			isa = PBXNativeTarget;
+			buildConfigurationList = E09D16962F44CA20009C51A3 /* Build configuration list for PBXNativeTarget "EXO-iOSUITests" */;
+			buildPhases = (
+				E09D16822F44CA20009C51A3 /* Sources */,
+				E09D16832F44CA20009C51A3 /* Frameworks */,
+				E09D16842F44CA20009C51A3 /* Resources */,
+			);
+			buildRules = (
+			);
+			dependencies = (
+				E09D16882F44CA20009C51A3 /* PBXTargetDependency */,
+			);
+			fileSystemSynchronizedGroups = (
+				E09D16892F44CA20009C51A3 /* EXO-iOSUITests */,
+			);
+			name = "EXO-iOSUITests";
+			packageProductDependencies = (
+			);
+			productName = "EXO-iOSUITests";
+			productReference = E09D16862F44CA20009C51A3 /* EXO-iOSUITests.xctest */;
+			productType = "com.apple.product-type.bundle.ui-testing";
+		};
+/* End PBXNativeTarget section */
+
+/* Begin PBXProject section */
+		E09D16672F44CA1E009C51A3 /* Project object */ = {
+			isa = PBXProject;
+			attributes = {
+				BuildIndependentTargetsInParallel = 1;
+				LastSwiftUpdateCheck = 2620;
+				LastUpgradeCheck = 2620;
+				TargetAttributes = {
+					E09D166E2F44CA1E009C51A3 = {
+						CreatedOnToolsVersion = 26.2;
+					};
+					E09D167B2F44CA20009C51A3 = {
+						CreatedOnToolsVersion = 26.2;
+						TestTargetID = E09D166E2F44CA1E009C51A3;
+					};
+					E09D16852F44CA20009C51A3 = {
+						CreatedOnToolsVersion = 26.2;
+						TestTargetID = E09D166E2F44CA1E009C51A3;
+					};
+				};
+			};
+			buildConfigurationList = E09D166A2F44CA1E009C51A3 /* Build configuration list for PBXProject "EXO-iOS" */;
+			developmentRegion = en;
+			hasScannedForEncodings = 0;
+			knownRegions = (
+				en,
+				Base,
+			);
+			mainGroup = E09D16662F44CA1E009C51A3;
+			minimizedProjectReferenceProxies = 1;
+			packageReferences = (
+				E09D17502F44F359009C51A3 /* XCRemoteSwiftPackageReference "mlx-swift-lm" */,
+			);
+			preferredProjectObjectVersion = 77;
+			productRefGroup = E09D16702F44CA1E009C51A3 /* Products */;
+			projectDirPath = "";
+			projectRoot = "";
+			targets = (
+				E09D166E2F44CA1E009C51A3 /* EXO-iOS */,
+				E09D167B2F44CA20009C51A3 /* EXO-iOSTests */,
+				E09D16852F44CA20009C51A3 /* EXO-iOSUITests */,
+			);
+		};
+/* End PBXProject section */
+
+/* Begin PBXResourcesBuildPhase section */
+		E09D166D2F44CA1E009C51A3 /* Resources */ = {
+			isa = PBXResourcesBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+		E09D167A2F44CA20009C51A3 /* Resources */ = {
+			isa = PBXResourcesBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+		E09D16842F44CA20009C51A3 /* Resources */ = {
+			isa = PBXResourcesBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+/* End PBXResourcesBuildPhase section */
+
+/* Begin PBXSourcesBuildPhase section */
+		E09D166B2F44CA1E009C51A3 /* Sources */ = {
+			isa = PBXSourcesBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+		E09D16782F44CA20009C51A3 /* Sources */ = {
+			isa = PBXSourcesBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+		E09D16822F44CA20009C51A3 /* Sources */ = {
+			isa = PBXSourcesBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+/* End PBXSourcesBuildPhase section */
+
+/* Begin PBXTargetDependency section */
+		E09D167E2F44CA20009C51A3 /* PBXTargetDependency */ = {
+			isa = PBXTargetDependency;
+			target = E09D166E2F44CA1E009C51A3 /* EXO-iOS */;
+			targetProxy = E09D167D2F44CA20009C51A3 /* PBXContainerItemProxy */;
+		};
+		E09D16882F44CA20009C51A3 /* PBXTargetDependency */ = {
+			isa = PBXTargetDependency;
+			target = E09D166E2F44CA1E009C51A3 /* EXO-iOS */;
+			targetProxy = E09D16872F44CA20009C51A3 /* PBXContainerItemProxy */;
+		};
+/* End PBXTargetDependency section */
+
+/* Begin XCBuildConfiguration section */
+		E09D168E2F44CA20009C51A3 /* Debug */ = {
+			isa = XCBuildConfiguration;
+			buildSettings = {
+				ALWAYS_SEARCH_USER_PATHS = NO;
+				ASSETCATALOG_COMPILER_GENERATE_SWIFT_ASSET_SYMBOL_EXTENSIONS = YES;
+				CLANG_ANALYZER_NONNULL = YES;
+				CLANG_ANALYZER_NUMBER_OBJECT_CONVERSION = YES_AGGRESSIVE;
+				CLANG_CXX_LANGUAGE_STANDARD = "gnu++20";
+				CLANG_ENABLE_MODULES = YES;
+				CLANG_ENABLE_OBJC_ARC = YES;
+				CLANG_ENABLE_OBJC_WEAK = YES;
+				CLANG_WARN_BLOCK_CAPTURE_AUTORELEASING = YES;
+				CLANG_WARN_BOOL_CONVERSION = YES;
+				CLANG_WARN_COMMA = YES;
+				CLANG_WARN_CONSTANT_CONVERSION = YES;
+				CLANG_WARN_DEPRECATED_OBJC_IMPLEMENTATIONS = YES;
+				CLANG_WARN_DIRECT_OBJC_ISA_USAGE = YES_ERROR;
+				CLANG_WARN_DOCUMENTATION_COMMENTS = YES;
+				CLANG_WARN_EMPTY_BODY = YES;
+				CLANG_WARN_ENUM_CONVERSION = YES;
+				CLANG_WARN_INFINITE_RECURSION = YES;
+				CLANG_WARN_INT_CONVERSION = YES;
+				CLANG_WARN_NON_LITERAL_NULL_CONVERSION = YES;
+				CLANG_WARN_OBJC_IMPLICIT_RETAIN_SELF = YES;
+				CLANG_WARN_OBJC_LITERAL_CONVERSION = YES;
+				CLANG_WARN_OBJC_ROOT_CLASS = YES_ERROR;
+				CLANG_WARN_QUOTED_INCLUDE_IN_FRAMEWORK_HEADER = YES;
+				CLANG_WARN_RANGE_LOOP_ANALYSIS = YES;
+				CLANG_WARN_STRICT_PROTOTYPES = YES;
+				CLANG_WARN_SUSPICIOUS_MOVE = YES;
+				CLANG_WARN_UNGUARDED_AVAILABILITY = YES_AGGRESSIVE;
+				CLANG_WARN_UNREACHABLE_CODE = YES;
+				CLANG_WARN__DUPLICATE_METHOD_MATCH = YES;
+				COPY_PHASE_STRIP = NO;
+				DEBUG_INFORMATION_FORMAT = dwarf;
+				ENABLE_STRICT_OBJC_MSGSEND = YES;
+				ENABLE_TESTABILITY = YES;
+				ENABLE_USER_SCRIPT_SANDBOXING = YES;
+				GCC_C_LANGUAGE_STANDARD = gnu17;
+				GCC_DYNAMIC_NO_PIC = NO;
+				GCC_NO_COMMON_BLOCKS = YES;
+				GCC_OPTIMIZATION_LEVEL = 0;
+				GCC_PREPROCESSOR_DEFINITIONS = (
+					"DEBUG=1",
+					"$(inherited)",
+				);
+				GCC_WARN_64_TO_32_BIT_CONVERSION = YES;
+				GCC_WARN_ABOUT_RETURN_TYPE = YES_ERROR;
+				GCC_WARN_UNDECLARED_SELECTOR = YES;
+				GCC_WARN_UNINITIALIZED_AUTOS = YES_AGGRESSIVE;
+				GCC_WARN_UNUSED_FUNCTION = YES;
+				GCC_WARN_UNUSED_VARIABLE = YES;
+				IPHONEOS_DEPLOYMENT_TARGET = 26.2;
+				LOCALIZATION_PREFERS_STRING_CATALOGS = YES;
+				MTL_ENABLE_DEBUG_INFO = INCLUDE_SOURCE;
+				MTL_FAST_MATH = YES;
+				ONLY_ACTIVE_ARCH = YES;
+				SDKROOT = iphoneos;
+				SWIFT_ACTIVE_COMPILATION_CONDITIONS = "DEBUG $(inherited)";
+				SWIFT_OPTIMIZATION_LEVEL = "-Onone";
+			};
+			name = Debug;
+		};
+		E09D168F2F44CA20009C51A3 /* Release */ = {
+			isa = XCBuildConfiguration;
+			buildSettings = {
+				ALWAYS_SEARCH_USER_PATHS = NO;
+				ASSETCATALOG_COMPILER_GENERATE_SWIFT_ASSET_SYMBOL_EXTENSIONS = YES;
+				CLANG_ANALYZER_NONNULL = YES;
+				CLANG_ANALYZER_NUMBER_OBJECT_CONVERSION = YES_AGGRESSIVE;
+				CLANG_CXX_LANGUAGE_STANDARD = "gnu++20";
+				CLANG_ENABLE_MODULES = YES;
+				CLANG_ENABLE_OBJC_ARC = YES;
+				CLANG_ENABLE_OBJC_WEAK = YES;
+				CLANG_WARN_BLOCK_CAPTURE_AUTORELEASING = YES;
+				CLANG_WARN_BOOL_CONVERSION = YES;
+				CLANG_WARN_COMMA = YES;
+				CLANG_WARN_CONSTANT_CONVERSION = YES;
+				CLANG_WARN_DEPRECATED_OBJC_IMPLEMENTATIONS = YES;
+				CLANG_WARN_DIRECT_OBJC_ISA_USAGE = YES_ERROR;
+				CLANG_WARN_DOCUMENTATION_COMMENTS = YES;
+				CLANG_WARN_EMPTY_BODY = YES;
+				CLANG_WARN_ENUM_CONVERSION = YES;
+				CLANG_WARN_INFINITE_RECURSION = YES;
+				CLANG_WARN_INT_CONVERSION = YES;
+				CLANG_WARN_NON_LITERAL_NULL_CONVERSION = YES;
+				CLANG_WARN_OBJC_IMPLICIT_RETAIN_SELF = YES;
+				CLANG_WARN_OBJC_LITERAL_CONVERSION = YES;
+				CLANG_WARN_OBJC_ROOT_CLASS = YES_ERROR;
+				CLANG_WARN_QUOTED_INCLUDE_IN_FRAMEWORK_HEADER = YES;
+				CLANG_WARN_RANGE_LOOP_ANALYSIS = YES;
+				CLANG_WARN_STRICT_PROTOTYPES = YES;
+				CLANG_WARN_SUSPICIOUS_MOVE = YES;
+				CLANG_WARN_UNGUARDED_AVAILABILITY = YES_AGGRESSIVE;
+				CLANG_WARN_UNREACHABLE_CODE = YES;
+				CLANG_WARN__DUPLICATE_METHOD_MATCH = YES;
+				COPY_PHASE_STRIP = NO;
+				DEBUG_INFORMATION_FORMAT = "dwarf-with-dsym";
+				ENABLE_NS_ASSERTIONS = NO;
+				ENABLE_STRICT_OBJC_MSGSEND = YES;
+				ENABLE_USER_SCRIPT_SANDBOXING = YES;
+				GCC_C_LANGUAGE_STANDARD = gnu17;
+				GCC_NO_COMMON_BLOCKS = YES;
+				GCC_WARN_64_TO_32_BIT_CONVERSION = YES;
+				GCC_WARN_ABOUT_RETURN_TYPE = YES_ERROR;
+				GCC_WARN_UNDECLARED_SELECTOR = YES;
+				GCC_WARN_UNINITIALIZED_AUTOS = YES_AGGRESSIVE;
+				GCC_WARN_UNUSED_FUNCTION = YES;
+				GCC_WARN_UNUSED_VARIABLE = YES;
+				IPHONEOS_DEPLOYMENT_TARGET = 26.2;
+				LOCALIZATION_PREFERS_STRING_CATALOGS = YES;
+				MTL_ENABLE_DEBUG_INFO = NO;
+				MTL_FAST_MATH = YES;
+				SDKROOT = iphoneos;
+				SWIFT_COMPILATION_MODE = wholemodule;
+				VALIDATE_PRODUCT = YES;
+			};
+			name = Release;
+		};
+		E09D16912F44CA20009C51A3 /* Debug */ = {
+			isa = XCBuildConfiguration;
+			buildSettings = {
+				ASSETCATALOG_COMPILER_APPICON_NAME = AppIcon;
+				ASSETCATALOG_COMPILER_GLOBAL_ACCENT_COLOR_NAME = AccentColor;
+				CODE_SIGN_STYLE = Automatic;
+				CURRENT_PROJECT_VERSION = 1;
+				DEVELOPMENT_TEAM = 3M3M67U93M;
+				ENABLE_PREVIEWS = YES;
+				GENERATE_INFOPLIST_FILE = YES;
+				INFOPLIST_FILE = "EXO-iOS/Info.plist";
+				INFOPLIST_KEY_UIApplicationSceneManifest_Generation = YES;
+				INFOPLIST_KEY_UIApplicationSupportsIndirectInputEvents = YES;
+				INFOPLIST_KEY_UILaunchScreen_Generation = YES;
+				INFOPLIST_KEY_UISupportedInterfaceOrientations_iPad = "UIInterfaceOrientationPortrait UIInterfaceOrientationPortraitUpsideDown UIInterfaceOrientationLandscapeLeft UIInterfaceOrientationLandscapeRight";
+				INFOPLIST_KEY_UISupportedInterfaceOrientations_iPhone = "UIInterfaceOrientationPortrait UIInterfaceOrientationLandscapeLeft UIInterfaceOrientationLandscapeRight";
+				LD_RUNPATH_SEARCH_PATHS = (
+					"$(inherited)",
+					"@executable_path/Frameworks",
+				);
+				MARKETING_VERSION = 1.0;
+				PRODUCT_BUNDLE_IDENTIFIER = "com.exo.EXO-iOS";
+				PRODUCT_NAME = "$(TARGET_NAME)";
+				STRING_CATALOG_GENERATE_SYMBOLS = YES;
+				SWIFT_APPROACHABLE_CONCURRENCY = YES;
+				SWIFT_DEFAULT_ACTOR_ISOLATION = MainActor;
+				SWIFT_EMIT_LOC_STRINGS = YES;
+				SWIFT_UPCOMING_FEATURE_MEMBER_IMPORT_VISIBILITY = YES;
+				SWIFT_VERSION = 5.0;
+				TARGETED_DEVICE_FAMILY = "1,2";
+			};
+			name = Debug;
+		};
+		E09D16922F44CA20009C51A3 /* Release */ = {
+			isa = XCBuildConfiguration;
+			buildSettings = {
+				ASSETCATALOG_COMPILER_APPICON_NAME = AppIcon;
+				ASSETCATALOG_COMPILER_GLOBAL_ACCENT_COLOR_NAME = AccentColor;
+				CODE_SIGN_STYLE = Automatic;
+				CURRENT_PROJECT_VERSION = 1;
+				DEVELOPMENT_TEAM = 3M3M67U93M;
+				ENABLE_PREVIEWS = YES;
+				GENERATE_INFOPLIST_FILE = YES;
+				INFOPLIST_FILE = "EXO-iOS/Info.plist";
+				INFOPLIST_KEY_UIApplicationSceneManifest_Generation = YES;
+				INFOPLIST_KEY_UIApplicationSupportsIndirectInputEvents = YES;
+				INFOPLIST_KEY_UILaunchScreen_Generation = YES;
+				INFOPLIST_KEY_UISupportedInterfaceOrientations_iPad = "UIInterfaceOrientationPortrait UIInterfaceOrientationPortraitUpsideDown UIInterfaceOrientationLandscapeLeft UIInterfaceOrientationLandscapeRight";
+				INFOPLIST_KEY_UISupportedInterfaceOrientations_iPhone = "UIInterfaceOrientationPortrait UIInterfaceOrientationLandscapeLeft UIInterfaceOrientationLandscapeRight";
+				LD_RUNPATH_SEARCH_PATHS = (
+					"$(inherited)",
+					"@executable_path/Frameworks",
+				);
+				MARKETING_VERSION = 1.0;
+				PRODUCT_BUNDLE_IDENTIFIER = "com.exo.EXO-iOS";
+				PRODUCT_NAME = "$(TARGET_NAME)";
+				STRING_CATALOG_GENERATE_SYMBOLS = YES;
+				SWIFT_APPROACHABLE_CONCURRENCY = YES;
+				SWIFT_DEFAULT_ACTOR_ISOLATION = MainActor;
+				SWIFT_EMIT_LOC_STRINGS = YES;
+				SWIFT_UPCOMING_FEATURE_MEMBER_IMPORT_VISIBILITY = YES;
+				SWIFT_VERSION = 5.0;
+				TARGETED_DEVICE_FAMILY = "1,2";
+			};
+			name = Release;
+		};
+		E09D16942F44CA20009C51A3 /* Debug */ = {
+			isa = XCBuildConfiguration;
+			buildSettings = {
+				BUNDLE_LOADER = "$(TEST_HOST)";
+				CODE_SIGN_STYLE = Automatic;
+				CURRENT_PROJECT_VERSION = 1;
+				GENERATE_INFOPLIST_FILE = YES;
+				IPHONEOS_DEPLOYMENT_TARGET = 26.2;
+				MARKETING_VERSION = 1.0;
+				PRODUCT_BUNDLE_IDENTIFIER = "com.exo.EXO-iOSTests";
+				PRODUCT_NAME = "$(TARGET_NAME)";
+				STRING_CATALOG_GENERATE_SYMBOLS = NO;
+				SWIFT_APPROACHABLE_CONCURRENCY = YES;
+				SWIFT_EMIT_LOC_STRINGS = NO;
+				SWIFT_UPCOMING_FEATURE_MEMBER_IMPORT_VISIBILITY = YES;
+				SWIFT_VERSION = 5.0;
+				TARGETED_DEVICE_FAMILY = "1,2";
+				TEST_HOST = "$(BUILT_PRODUCTS_DIR)/EXO-iOS.app/$(BUNDLE_EXECUTABLE_FOLDER_PATH)/EXO-iOS";
+			};
+			name = Debug;
+		};
+		E09D16952F44CA20009C51A3 /* Release */ = {
+			isa = XCBuildConfiguration;
+			buildSettings = {
+				BUNDLE_LOADER = "$(TEST_HOST)";
+				CODE_SIGN_STYLE = Automatic;
+				CURRENT_PROJECT_VERSION = 1;
+				GENERATE_INFOPLIST_FILE = YES;
+				IPHONEOS_DEPLOYMENT_TARGET = 26.2;
+				MARKETING_VERSION = 1.0;
+				PRODUCT_BUNDLE_IDENTIFIER = "com.exo.EXO-iOSTests";
+				PRODUCT_NAME = "$(TARGET_NAME)";
+				STRING_CATALOG_GENERATE_SYMBOLS = NO;
+				SWIFT_APPROACHABLE_CONCURRENCY = YES;
+				SWIFT_EMIT_LOC_STRINGS = NO;
+				SWIFT_UPCOMING_FEATURE_MEMBER_IMPORT_VISIBILITY = YES;
+				SWIFT_VERSION = 5.0;
+				TARGETED_DEVICE_FAMILY = "1,2";
+				TEST_HOST = "$(BUILT_PRODUCTS_DIR)/EXO-iOS.app/$(BUNDLE_EXECUTABLE_FOLDER_PATH)/EXO-iOS";
+			};
+			name = Release;
+		};
+		E09D16972F44CA20009C51A3 /* Debug */ = {
+			isa = XCBuildConfiguration;
+			buildSettings = {
+				CODE_SIGN_STYLE = Automatic;
+				CURRENT_PROJECT_VERSION = 1;
+				GENERATE_INFOPLIST_FILE = YES;
+				MARKETING_VERSION = 1.0;
+				PRODUCT_BUNDLE_IDENTIFIER = "com.exo.EXO-iOSUITests";
+				PRODUCT_NAME = "$(TARGET_NAME)";
+				STRING_CATALOG_GENERATE_SYMBOLS = NO;
+				SWIFT_APPROACHABLE_CONCURRENCY = YES;
+				SWIFT_EMIT_LOC_STRINGS = NO;
+				SWIFT_UPCOMING_FEATURE_MEMBER_IMPORT_VISIBILITY = YES;
+				SWIFT_VERSION = 5.0;
+				TARGETED_DEVICE_FAMILY = "1,2";
+				TEST_TARGET_NAME = "EXO-iOS";
+			};
+			name = Debug;
+		};
+		E09D16982F44CA20009C51A3 /* Release */ = {
+			isa = XCBuildConfiguration;
+			buildSettings = {
+				CODE_SIGN_STYLE = Automatic;
+				CURRENT_PROJECT_VERSION = 1;
+				GENERATE_INFOPLIST_FILE = YES;
+				MARKETING_VERSION = 1.0;
+				PRODUCT_BUNDLE_IDENTIFIER = "com.exo.EXO-iOSUITests";
+				PRODUCT_NAME = "$(TARGET_NAME)";
+				STRING_CATALOG_GENERATE_SYMBOLS = NO;
+				SWIFT_APPROACHABLE_CONCURRENCY = YES;
+				SWIFT_EMIT_LOC_STRINGS = NO;
+				SWIFT_UPCOMING_FEATURE_MEMBER_IMPORT_VISIBILITY = YES;
+				SWIFT_VERSION = 5.0;
+				TARGETED_DEVICE_FAMILY = "1,2";
+				TEST_TARGET_NAME = "EXO-iOS";
+			};
+			name = Release;
+		};
+/* End XCBuildConfiguration section */
+
+/* Begin XCConfigurationList section */
+		E09D166A2F44CA1E009C51A3 /* Build configuration list for PBXProject "EXO-iOS" */ = {
+			isa = XCConfigurationList;
+			buildConfigurations = (
+				E09D168E2F44CA20009C51A3 /* Debug */,
+				E09D168F2F44CA20009C51A3 /* Release */,
+			);
+			defaultConfigurationIsVisible = 0;
+			defaultConfigurationName = Release;
+		};
+		E09D16902F44CA20009C51A3 /* Build configuration list for PBXNativeTarget "EXO-iOS" */ = {
+			isa = XCConfigurationList;
+			buildConfigurations = (
+				E09D16912F44CA20009C51A3 /* Debug */,
+				E09D16922F44CA20009C51A3 /* Release */,
+			);
+			defaultConfigurationIsVisible = 0;
+			defaultConfigurationName = Release;
+		};
+		E09D16932F44CA20009C51A3 /* Build configuration list for PBXNativeTarget "EXO-iOSTests" */ = {
+			isa = XCConfigurationList;
+			buildConfigurations = (
+				E09D16942F44CA20009C51A3 /* Debug */,
+				E09D16952F44CA20009C51A3 /* Release */,
+			);
+			defaultConfigurationIsVisible = 0;
+			defaultConfigurationName = Release;
+		};
+		E09D16962F44CA20009C51A3 /* Build configuration list for PBXNativeTarget "EXO-iOSUITests" */ = {
+			isa = XCConfigurationList;
+			buildConfigurations = (
+				E09D16972F44CA20009C51A3 /* Debug */,
+				E09D16982F44CA20009C51A3 /* Release */,
+			);
+			defaultConfigurationIsVisible = 0;
+			defaultConfigurationName = Release;
+		};
+/* End XCConfigurationList section */
+
+/* Begin XCRemoteSwiftPackageReference section */
+		E09D17502F44F359009C51A3 /* XCRemoteSwiftPackageReference "mlx-swift-lm" */ = {
+			isa = XCRemoteSwiftPackageReference;
+			repositoryURL = "https://github.com/ml-explore/mlx-swift-lm";
+			requirement = {
+				kind = upToNextMajorVersion;
+				minimumVersion = 2.30.3;
+			};
+		};
+/* End XCRemoteSwiftPackageReference section */
+
+/* Begin XCSwiftPackageProductDependency section */
+		E09D17512F44F359009C51A3 /* MLXLLM */ = {
+			isa = XCSwiftPackageProductDependency;
+			package = E09D17502F44F359009C51A3 /* XCRemoteSwiftPackageReference "mlx-swift-lm" */;
+			productName = MLXLLM;
+		};
+		E09D17532F44F359009C51A3 /* MLXLMCommon */ = {
+			isa = XCSwiftPackageProductDependency;
+			package = E09D17502F44F359009C51A3 /* XCRemoteSwiftPackageReference "mlx-swift-lm" */;
+			productName = MLXLMCommon;
+		};
+/* End XCSwiftPackageProductDependency section */
+	};
+	rootObject = E09D16672F44CA1E009C51A3 /* Project object */;
+}
@@ -0,0 +1,7 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<Workspace
+   version = "1.0">
+   <FileRef
+      location = "self:">
+   </FileRef>
+</Workspace>
@@ -0,0 +1,60 @@
+{
+  "originHash" : "facc0ac7c70363ea20f6cd1235de91dea6b06f0d00190946045a6c8ae753abc2",
+  "pins" : [
+    {
+      "identity" : "mlx-swift",
+      "kind" : "remoteSourceControl",
+      "location" : "https://github.com/ml-explore/mlx-swift",
+      "state" : {
+        "revision" : "6ba4827fb82c97d012eec9ab4b2de21f85c3b33d",
+        "version" : "0.30.6"
+      }
+    },
+    {
+      "identity" : "mlx-swift-lm",
+      "kind" : "remoteSourceControl",
+      "location" : "https://github.com/ml-explore/mlx-swift-lm",
+      "state" : {
+        "revision" : "360c5052b81cc154b04ee0933597a4ad6db4b8ae",
+        "version" : "2.30.3"
+      }
+    },
+    {
+      "identity" : "swift-collections",
+      "kind" : "remoteSourceControl",
+      "location" : "https://github.com/apple/swift-collections.git",
+      "state" : {
+        "revision" : "7b847a3b7008b2dc2f47ca3110d8c782fb2e5c7e",
+        "version" : "1.3.0"
+      }
+    },
+    {
+      "identity" : "swift-jinja",
+      "kind" : "remoteSourceControl",
+      "location" : "https://github.com/huggingface/swift-jinja.git",
+      "state" : {
+        "revision" : "d81197f35f41445bc10e94600795e68c6f5e94b0",
+        "version" : "2.3.1"
+      }
+    },
+    {
+      "identity" : "swift-numerics",
+      "kind" : "remoteSourceControl",
+      "location" : "https://github.com/apple/swift-numerics",
+      "state" : {
+        "revision" : "0c0290ff6b24942dadb83a929ffaaa1481df04a2",
+        "version" : "1.1.1"
+      }
+    },
+    {
+      "identity" : "swift-transformers",
+      "kind" : "remoteSourceControl",
+      "location" : "https://github.com/huggingface/swift-transformers",
+      "state" : {
+        "revision" : "573e5c9036c2f136b3a8a071da8e8907322403d0",
+        "version" : "1.1.6"
+      }
+    }
+  ],
+  "version" : 3
+}
@@ -0,0 +1,20 @@
+{
+  "colors" : [
+    {
+      "color" : {
+        "color-space" : "srgb",
+        "components" : {
+          "alpha" : "1.000",
+          "blue" : "0x00",
+          "green" : "0xD7",
+          "red" : "0xFF"
+        }
+      },
+      "idiom" : "universal"
+    }
+  ],
+  "info" : {
+    "author" : "xcode",
+    "version" : 1
+  }
+}
@@ -0,0 +1,38 @@
+{
+  "images" : [
+    {
+      "filename" : "AppIcon.png",
+      "idiom" : "universal",
+      "platform" : "ios",
+      "size" : "1024x1024"
+    },
+    {
+      "appearances" : [
+        {
+          "appearance" : "luminosity",
+          "value" : "dark"
+        }
+      ],
+      "filename" : "AppIcon.png",
+      "idiom" : "universal",
+      "platform" : "ios",
+      "size" : "1024x1024"
+    },
+    {
+      "appearances" : [
+        {
+          "appearance" : "luminosity",
+          "value" : "tinted"
+        }
+      ],
+      "filename" : "AppIcon.png",
+      "idiom" : "universal",
+      "platform" : "ios",
+      "size" : "1024x1024"
+    }
+  ],
+  "info" : {
+    "author" : "xcode",
+    "version" : 1
+  }
+}
@@ -0,0 +1,6 @@
+{
+  "info" : {
+    "author" : "xcode",
+    "version" : 1
+  }
+}
@@ -0,0 +1,21 @@
+{
+  "images" : [
+    {
+      "filename" : "exo-logo.png",
+      "idiom" : "universal",
+      "scale" : "1x"
+    },
+    {
+      "idiom" : "universal",
+      "scale" : "2x"
+    },
+    {
+      "idiom" : "universal",
+      "scale" : "3x"
+    }
+  ],
+  "info" : {
+    "author" : "xcode",
+    "version" : 1
+  }
+}
@@ -0,0 +1,8 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+	<key>com.apple.developer.kernel.increased-memory-limit</key>
+	<true/>
+</dict>
+</plist>
@@ -0,0 +1,67 @@
+import SwiftUI
+import UIKit
+
+@main
+struct EXO_iOSApp: App {
+    @State private var clusterService = ClusterService()
+    @State private var discoveryService = DiscoveryService()
+    @State private var localInferenceService = LocalInferenceService()
+    @State private var chatService: ChatService?
+
+    init() {
+        let darkGray = UIColor(red: 0x1F / 255.0, green: 0x1F / 255.0, blue: 0x1F / 255.0, alpha: 1)
+        let yellow = UIColor(red: 0xFF / 255.0, green: 0xD7 / 255.0, blue: 0x00 / 255.0, alpha: 1)
+
+        let navAppearance = UINavigationBarAppearance()
+        navAppearance.configureWithOpaqueBackground()
+        navAppearance.backgroundColor = darkGray
+        navAppearance.titleTextAttributes = [
+            .foregroundColor: yellow,
+            .font: UIFont.monospacedSystemFont(ofSize: 17, weight: .semibold),
+        ]
+        navAppearance.largeTitleTextAttributes = [
+            .foregroundColor: yellow,
+            .font: UIFont.monospacedSystemFont(ofSize: 34, weight: .bold),
+        ]
+
+        UINavigationBar.appearance().standardAppearance = navAppearance
+        UINavigationBar.appearance().compactAppearance = navAppearance
+        UINavigationBar.appearance().scrollEdgeAppearance = navAppearance
+        UINavigationBar.appearance().tintColor = yellow
+    }
+
+    var body: some Scene {
+        WindowGroup {
+            if let chatService {
+                RootView()
+                    .environment(clusterService)
+                    .environment(discoveryService)
+                    .environment(chatService)
+                    .environment(localInferenceService)
+                    .preferredColorScheme(.dark)
+                    .task {
+                        await clusterService.attemptAutoReconnect()
+                        discoveryService.startBrowsing()
+                        await localInferenceService.prepareModel()
+                    }
+                    .onChange(of: discoveryService.discoveredClusters) { _, clusters in
+                        guard !clusterService.isConnected,
+                            case .disconnected = clusterService.connectionState,
+                            let first = clusters.first
+                        else { return }
+                        Task {
+                            await clusterService.connectToDiscoveredCluster(
+                                first, using: discoveryService)
+                        }
+                    }
+            } else {
+                Color.exoBlack.onAppear {
+                    chatService = ChatService(
+                        clusterService: clusterService,
+                        localInferenceService: localInferenceService
+                    )
+                }
+            }
+        }
+    }
+}
@@ -0,0 +1,19 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+	<key>UIUserInterfaceStyle</key>
+	<string>Dark</string>
+	<key>CFBundleDisplayName</key>
+	<string>EXO</string>
+	<key>NSLocalNetworkUsageDescription</key>
+	<string>EXO needs local network access to connect to your EXO cluster.</string>
+	<key>NSBonjourServices</key>
+	<array>
+		<string>_exo._tcp</string>
+		<string>_p2p._tcp</string>
+		<string>_p2p._udp</string>
+		<string>_libp2p._udp</string>
+	</array>
+</dict>
+</plist>
@@ -0,0 +1,129 @@
+import Foundation
+
+// MARK: - Request
+
+struct ChatCompletionRequest: Encodable {
+    let model: String
+    let messages: [ChatCompletionMessageParam]
+    let stream: Bool
+    let maxTokens: Int?
+    let temperature: Double?
+
+    enum CodingKeys: String, CodingKey {
+        case model, messages, stream, temperature
+        case maxTokens = "max_tokens"
+    }
+}
+
+struct ChatCompletionMessageParam: Encodable {
+    let role: String
+    let content: String
+}
+
+// MARK: - Streaming Response
+
+struct ChatCompletionChunk: Decodable {
+    let id: String
+    let model: String?
+    let choices: [StreamingChoice]
+    let usage: ChunkUsage?
+
+    init(id: String, model: String?, choices: [StreamingChoice], usage: ChunkUsage?) {
+        self.id = id
+        self.model = model
+        self.choices = choices
+        self.usage = usage
+    }
+}
+
+struct StreamingChoice: Decodable {
+    let index: Int
+    let delta: Delta
+    let finishReason: String?
+
+    enum CodingKeys: String, CodingKey {
+        case index, delta
+        case finishReason = "finish_reason"
+    }
+
+    init(index: Int, delta: Delta, finishReason: String?) {
+        self.index = index
+        self.delta = delta
+        self.finishReason = finishReason
+    }
+}
+
+struct Delta: Decodable {
+    let role: String?
+    let content: String?
+
+    init(role: String?, content: String?) {
+        self.role = role
+        self.content = content
+    }
+}
+
+struct ChunkUsage: Decodable {
+    let promptTokens: Int?
+    let completionTokens: Int?
+    let totalTokens: Int?
+
+    enum CodingKeys: String, CodingKey {
+        case promptTokens = "prompt_tokens"
+        case completionTokens = "completion_tokens"
+        case totalTokens = "total_tokens"
+    }
+
+    init(promptTokens: Int?, completionTokens: Int?, totalTokens: Int?) {
+        self.promptTokens = promptTokens
+        self.completionTokens = completionTokens
+        self.totalTokens = totalTokens
+    }
+}
+
+// MARK: - Non-Streaming Response
+
+struct ChatCompletionResponse: Decodable {
+    let id: String
+    let model: String?
+    let choices: [ResponseChoice]
+}
+
+struct ResponseChoice: Decodable {
+    let index: Int
+    let message: ResponseMessage
+    let finishReason: String?
+
+    enum CodingKeys: String, CodingKey {
+        case index, message
+        case finishReason = "finish_reason"
+    }
+}
+
+struct ResponseMessage: Decodable {
+    let role: String?
+    let content: String?
+}
+
+// MARK: - Models List
+
+struct ModelListResponse: Decodable {
+    let data: [ModelInfo]
+}
+
+struct ModelInfo: Decodable, Identifiable {
+    let id: String
+    let name: String?
+}
+
+// MARK: - Error
+
+struct APIErrorResponse: Decodable {
+    let error: APIErrorInfo
+}
+
+struct APIErrorInfo: Decodable {
+    let message: String
+    let type: String?
+    let code: Int?
+}
@@ -0,0 +1,26 @@
+import Foundation
+
+struct ChatMessage: Identifiable, Equatable {
+    let id: UUID
+    let role: Role
+    var content: String
+    let timestamp: Date
+    var isStreaming: Bool
+
+    enum Role: String, Codable {
+        case user
+        case assistant
+        case system
+    }
+
+    init(
+        id: UUID = UUID(), role: Role, content: String, timestamp: Date = Date(),
+        isStreaming: Bool = false
+    ) {
+        self.id = id
+        self.role = role
+        self.content = content
+        self.timestamp = timestamp
+        self.isStreaming = isStreaming
+    }
+}
@@ -0,0 +1,11 @@
+import Foundation
+
+struct ConnectionInfo: Codable, Equatable {
+    let host: String
+    let port: Int
+    let nodeId: String?
+
+    var baseURL: URL { URL(string: "http://\(host):\(port)")! }
+
+    static let defaultPort = 52415
+}
@@ -0,0 +1,34 @@
+import Foundation
+
+struct Conversation: Identifiable, Codable, Equatable {
+    let id: UUID
+    var title: String
+    var messages: [StoredMessage]
+    var modelId: String?
+    let createdAt: Date
+
+    init(
+        id: UUID = UUID(), title: String = "New Chat", messages: [StoredMessage] = [],
+        modelId: String? = nil, createdAt: Date = Date()
+    ) {
+        self.id = id
+        self.title = title
+        self.messages = messages
+        self.modelId = modelId
+        self.createdAt = createdAt
+    }
+}
+
+struct StoredMessage: Identifiable, Codable, Equatable {
+    let id: UUID
+    let role: String
+    var content: String
+    let timestamp: Date
+
+    init(id: UUID = UUID(), role: String, content: String, timestamp: Date = Date()) {
+        self.id = id
+        self.role = role
+        self.content = content
+        self.timestamp = timestamp
+    }
+}
@@ -0,0 +1,227 @@
+import Foundation
+
+@Observable
+@MainActor
+final class ChatService {
+    var conversations: [Conversation] = []
+    var activeConversationId: UUID?
+    private(set) var isGenerating: Bool = false
+    private var currentGenerationTask: Task<Void, Never>?
+
+    private let clusterService: ClusterService
+    private let localInferenceService: LocalInferenceService
+
+    var canSendMessage: Bool {
+        clusterService.isConnected || localInferenceService.isAvailable
+    }
+
+    var activeConversation: Conversation? {
+        guard let id = activeConversationId else { return nil }
+        return conversations.first { $0.id == id }
+    }
+
+    var activeMessages: [ChatMessage] {
+        guard let conversation = activeConversation else { return [] }
+        return conversation.messages.map { stored in
+            ChatMessage(
+                id: stored.id,
+                role: ChatMessage.Role(rawValue: stored.role) ?? .user,
+                content: stored.content,
+                timestamp: stored.timestamp
+            )
+        }
+    }
+
+    init(clusterService: ClusterService, localInferenceService: LocalInferenceService) {
+        self.clusterService = clusterService
+        self.localInferenceService = localInferenceService
+        loadConversations()
+    }
+
+    // MARK: - Conversation Management
+
+    func createConversation(modelId: String? = nil) {
+        let conversation = Conversation(
+            modelId: modelId ?? clusterService.availableModels.first?.id)
+        conversations.insert(conversation, at: 0)
+        activeConversationId = conversation.id
+        saveConversations()
+    }
+
+    func deleteConversation(id: UUID) {
+        conversations.removeAll { $0.id == id }
+        if activeConversationId == id {
+            activeConversationId = conversations.first?.id
+        }
+        saveConversations()
+    }
+
+    func setActiveConversation(id: UUID) {
+        activeConversationId = id
+    }
+
+    func setModelForActiveConversation(_ modelId: String) {
+        guard let index = conversations.firstIndex(where: { $0.id == activeConversationId }) else {
+            return
+        }
+        conversations[index].modelId = modelId
+        saveConversations()
+    }
+
+    // MARK: - Messaging
+
+    func sendMessage(_ text: String) {
+        guard !text.trimmingCharacters(in: .whitespacesAndNewlines).isEmpty else { return }
+
+        if activeConversation == nil {
+            createConversation()
+        }
+
+        guard let index = conversations.firstIndex(where: { $0.id == activeConversationId }) else {
+            return
+        }
+
+        let userMessage = StoredMessage(role: "user", content: text)
+        conversations[index].messages.append(userMessage)
+
+        if conversations[index].title == "New Chat" {
+            let preview = String(text.prefix(40))
+            conversations[index].title = preview + (text.count > 40 ? "..." : "")
+        }
+
+        let modelId: String
+        if clusterService.isConnected {
+            guard
+                let clusterId = conversations[index].modelId
+                    ?? clusterService.availableModels.first?.id
+            else {
+                let errorMessage = StoredMessage(
+                    role: "assistant", content: "No model selected. Please select a model first.")
+                conversations[index].messages.append(errorMessage)
+                saveConversations()
+                return
+            }
+            modelId = clusterId
+        } else if localInferenceService.isAvailable {
+            modelId = localInferenceService.defaultModelId
+        } else {
+            let errorMessage = StoredMessage(
+                role: "assistant",
+                content: "Not connected to a cluster and local model is not available.")
+            conversations[index].messages.append(errorMessage)
+            saveConversations()
+            return
+        }
+
+        conversations[index].modelId = modelId
+
+        let assistantMessageId = UUID()
+        let assistantMessage = StoredMessage(
+            id: assistantMessageId, role: "assistant", content: "", timestamp: Date())
+        conversations[index].messages.append(assistantMessage)
+
+        let messagesForAPI = conversations[index].messages.dropLast().map { stored in
+            ChatCompletionMessageParam(role: stored.role, content: stored.content)
+        }
+
+        let request = ChatCompletionRequest(
+            model: modelId,
+            messages: Array(messagesForAPI),
+            stream: true,
+            maxTokens: 4096,
+            temperature: nil
+        )
+
+        let conversationId = conversations[index].id
+
+        isGenerating = true
+        currentGenerationTask = Task { [weak self] in
+            guard let self else { return }
+            await self.performStreaming(
+                request: request, conversationId: conversationId,
+                assistantMessageId: assistantMessageId)
+        }
+
+        saveConversations()
+    }
+
+    func cancelGeneration() {
+        currentGenerationTask?.cancel()
+        currentGenerationTask = nil
+        localInferenceService.cancelGeneration()
+        isGenerating = false
+    }
+
+    // MARK: - Streaming
+
+    private func performStreaming(
+        request: ChatCompletionRequest, conversationId: UUID, assistantMessageId: UUID
+    ) async {
+        defer {
+            isGenerating = false
+            currentGenerationTask = nil
+            saveConversations()
+        }
+
+        do {
+            let stream =
+                clusterService.isConnected
+                ? clusterService.streamChatCompletion(request: request)
+                : localInferenceService.streamChatCompletion(request: request)
+            for try await chunk in stream {
+                guard !Task.isCancelled else { return }
+                guard let content = chunk.choices.first?.delta.content, !content.isEmpty else {
+                    continue
+                }
+
+                if let convIndex = conversations.firstIndex(where: { $0.id == conversationId }),
+                    let msgIndex = conversations[convIndex].messages.firstIndex(where: {
+                        $0.id == assistantMessageId
+                    })
+                {
+                    conversations[convIndex].messages[msgIndex].content += content
+                }
+            }
+        } catch {
+            if !Task.isCancelled {
+                if let convIndex = conversations.firstIndex(where: { $0.id == conversationId }),
+                    let msgIndex = conversations[convIndex].messages.firstIndex(where: {
+                        $0.id == assistantMessageId
+                    })
+                {
+                    if conversations[convIndex].messages[msgIndex].content.isEmpty {
+                        conversations[convIndex].messages[msgIndex].content =
+                            "Error: \(error.localizedDescription)"
+                    }
+                }
+            }
+        }
+    }
+
+    // MARK: - Persistence
+
+    private static var storageURL: URL {
+        let documents = FileManager.default.urls(for: .documentDirectory, in: .userDomainMask)
+            .first!
+        return documents.appendingPathComponent("exo_conversations.json")
+    }
+
+    private func saveConversations() {
+        do {
+            let data = try JSONEncoder().encode(conversations)
+            try data.write(to: Self.storageURL, options: .atomic)
+        } catch {
+            // Save failed silently
+        }
+    }
+
+    private func loadConversations() {
+        do {
+            let data = try Data(contentsOf: Self.storageURL)
+            conversations = try JSONDecoder().decode([Conversation].self, from: data)
+            activeConversationId = conversations.first?.id
+        } catch {
+            conversations = []
+        }
+    }
+}
@@ -0,0 +1,246 @@
+import Foundation
+
+enum ConnectionState: Equatable {
+    case disconnected
+    case connecting
+    case connected(ConnectionInfo)
+}
+
+struct ModelOption: Identifiable, Equatable {
+    let id: String
+    let displayName: String
+}
+
+@Observable
+@MainActor
+final class ClusterService {
+    private(set) var connectionState: ConnectionState = .disconnected
+    private(set) var availableModels: [ModelOption] = []
+    private(set) var lastError: String?
+
+    private let session: URLSession
+    private let decoder: JSONDecoder
+    private var pollingTask: Task<Void, Never>?
+    private var heartbeatTask: Task<Void, Never>?
+
+    private static let connectionInfoKey = "exo_last_connection_info"
+
+    var isConnected: Bool {
+        if case .connected = connectionState { return true }
+        return false
+    }
+
+    var currentConnection: ConnectionInfo? {
+        if case .connected(let info) = connectionState { return info }
+        return nil
+    }
+
+    init(session: URLSession = .shared) {
+        self.session = session
+        let decoder = JSONDecoder()
+        self.decoder = decoder
+    }
+
+    // MARK: - Connection
+
+    func connect(to info: ConnectionInfo) async {
+        connectionState = .connecting
+        lastError = nil
+
+        do {
+            let url = info.baseURL.appendingPathComponent("node_id")
+            var request = URLRequest(url: url)
+            request.timeoutInterval = 5
+            request.cachePolicy = .reloadIgnoringLocalCacheData
+            let (_, response) = try await session.data(for: request)
+
+            guard let httpResponse = response as? HTTPURLResponse,
+                (200..<300).contains(httpResponse.statusCode)
+            else {
+                throw URLError(.badServerResponse)
+            }
+
+            connectionState = .connected(info)
+            persistConnection(info)
+            startPolling()
+            startHeartbeat()
+            await fetchModels(baseURL: info.baseURL)
+        } catch {
+            connectionState = .disconnected
+            lastError = "Could not connect to \(info.host):\(info.port)"
+        }
+    }
+
+    func connectToDiscoveredCluster(
+        _ cluster: DiscoveredCluster, using discoveryService: DiscoveryService
+    ) async {
+        guard case .disconnected = connectionState else { return }
+        connectionState = .connecting
+        lastError = nil
+
+        guard let info = await discoveryService.resolve(cluster) else {
+            connectionState = .disconnected
+            lastError = "Could not resolve \(cluster.name)"
+            return
+        }
+        connectionState = .disconnected  // reset so connect() can proceed
+        await connect(to: info)
+    }
+
+    func disconnect() {
+        stopPolling()
+        stopHeartbeat()
+        connectionState = .disconnected
+        availableModels = []
+        lastError = nil
+    }
+
+    func attemptAutoReconnect() async {
+        guard case .disconnected = connectionState,
+            let info = loadPersistedConnection()
+        else { return }
+        await connect(to: info)
+    }
+
+    // MARK: - Polling
+
+    private func startPolling(interval: TimeInterval = 2.0) {
+        stopPolling()
+        pollingTask = Task { [weak self] in
+            while !Task.isCancelled {
+                try? await Task.sleep(for: .seconds(interval))
+                guard let self, !Task.isCancelled else { return }
+                guard let connection = self.currentConnection else { return }
+                await self.fetchModels(baseURL: connection.baseURL)
+            }
+        }
+    }
+
+    private func stopPolling() {
+        pollingTask?.cancel()
+        pollingTask = nil
+    }
+
+    // MARK: - Heartbeat
+
+    private func startHeartbeat(interval: TimeInterval = 10.0) {
+        stopHeartbeat()
+        heartbeatTask = Task { [weak self] in
+            // Send immediately, then on interval
+            if let self, let connection = self.currentConnection {
+                await self.sendHeartbeat(baseURL: connection.baseURL)
+            }
+            while !Task.isCancelled {
+                try? await Task.sleep(for: .seconds(interval))
+                guard let self, !Task.isCancelled else { return }
+                guard let connection = self.currentConnection else { return }
+                await self.sendHeartbeat(baseURL: connection.baseURL)
+            }
+        }
+    }
+
+    private func stopHeartbeat() {
+        heartbeatTask?.cancel()
+        heartbeatTask = nil
+    }
+
+    private func sendHeartbeat(baseURL: URL) async {
+        do {
+            let deviceInfo = DeviceInfoService.gather()
+            let url = baseURL.appendingPathComponent("v1/lite_node/heartbeat")
+            var request = URLRequest(url: url)
+            request.httpMethod = "POST"
+            request.setValue("application/json", forHTTPHeaderField: "Content-Type")
+            request.timeoutInterval = 5
+            request.httpBody = try JSONEncoder().encode(deviceInfo)
+            let (_, response) = try await session.data(for: request)
+
+            if let httpResponse = response as? HTTPURLResponse,
+               !(200..<300).contains(httpResponse.statusCode) {
+                // Heartbeat failed silently — will retry on next interval
+            }
+        } catch {
+            // Heartbeat failed silently — will retry on next interval
+        }
+    }
+
+    // MARK: - API
+
+    private func fetchModels(baseURL: URL) async {
+        do {
+            let url = baseURL.appendingPathComponent("models")
+            var request = URLRequest(url: url)
+            request.cachePolicy = .reloadIgnoringLocalCacheData
+            let (data, response) = try await session.data(for: request)
+
+            guard let httpResponse = response as? HTTPURLResponse,
+                (200..<300).contains(httpResponse.statusCode)
+            else { return }
+
+            let list = try decoder.decode(ModelListResponse.self, from: data)
+            availableModels = list.data.map {
+                ModelOption(id: $0.id, displayName: $0.name ?? $0.id)
+            }
+        } catch {
+            // Models fetch failed silently — will retry on next poll
+        }
+    }
+
+    func streamChatCompletion(request body: ChatCompletionRequest) -> AsyncThrowingStream<
+        ChatCompletionChunk, Error
+    > {
+        AsyncThrowingStream { continuation in
+            let task = Task { [weak self] in
+                guard let self, let connection = self.currentConnection else {
+                    continuation.finish(throwing: URLError(.notConnectedToInternet))
+                    return
+                }
+
+                do {
+                    let url = connection.baseURL.appendingPathComponent("v1/chat/completions")
+                    var request = URLRequest(url: url)
+                    request.httpMethod = "POST"
+                    request.setValue("application/json", forHTTPHeaderField: "Content-Type")
+                    request.httpBody = try JSONEncoder().encode(body)
+
+                    let (bytes, response) = try await self.session.bytes(for: request)
+
+                    guard let httpResponse = response as? HTTPURLResponse,
+                        (200..<300).contains(httpResponse.statusCode)
+                    else {
+                        continuation.finish(throwing: URLError(.badServerResponse))
+                        return
+                    }
+
+                    let parser = SSEStreamParser<ChatCompletionChunk>(
+                        bytes: bytes, decoder: self.decoder)
+                    for try await chunk in parser {
+                        continuation.yield(chunk)
+                    }
+                    continuation.finish()
+                } catch {
+                    continuation.finish(throwing: error)
+                }
+            }
+
+            continuation.onTermination = { _ in
+                task.cancel()
+            }
+        }
+    }
+
+    // MARK: - Persistence
+
+    private func persistConnection(_ info: ConnectionInfo) {
+        if let data = try? JSONEncoder().encode(info) {
+            UserDefaults.standard.set(data, forKey: Self.connectionInfoKey)
+        }
+    }
+
+    private func loadPersistedConnection() -> ConnectionInfo? {
+        guard let data = UserDefaults.standard.data(forKey: Self.connectionInfoKey) else {
+            return nil
+        }
+        return try? JSONDecoder().decode(ConnectionInfo.self, from: data)
+    }
+}
@@ -0,0 +1,115 @@
+import Foundation
+import os
+import UIKit
+
+struct DeviceInfo: Encodable {
+    let nodeId: String
+    let model: String
+    let chip: String
+    let osVersion: String
+    let friendlyName: String
+    let ramTotal: Int
+    let ramAvailable: Int
+
+    enum CodingKeys: String, CodingKey {
+        case nodeId = "node_id"
+        case model
+        case chip
+        case osVersion = "os_version"
+        case friendlyName = "friendly_name"
+        case ramTotal = "ram_total"
+        case ramAvailable = "ram_available"
+    }
+}
+
+enum DeviceInfoService {
+    private static let liteNodeIdKey = "exo_lite_node_id"
+
+    static var liteNodeId: String {
+        if let existing = UserDefaults.standard.string(forKey: liteNodeIdKey) {
+            return existing
+        }
+        let newId = UUID().uuidString.lowercased()
+        UserDefaults.standard.set(newId, forKey: liteNodeIdKey)
+        return newId
+    }
+
+    static func gather() -> DeviceInfo {
+        let model = modelName()
+        let chip = chipName(for: model)
+
+        let totalRam = Int(ProcessInfo.processInfo.physicalMemory)
+        let availableRam = availableMemory()
+
+        return DeviceInfo(
+            nodeId: liteNodeId,
+            model: model,
+            chip: chip,
+            osVersion: UIDevice.current.systemVersion,
+            friendlyName: UIDevice.current.name,
+            ramTotal: totalRam,
+            ramAvailable: availableRam
+        )
+    }
+
+    // MARK: - Private
+
+    private static func modelName() -> String {
+        var systemInfo = utsname()
+        uname(&systemInfo)
+        let machine = withUnsafePointer(to: &systemInfo.machine) {
+            $0.withMemoryRebound(to: CChar.self, capacity: 1) {
+                String(cString: $0)
+            }
+        }
+        return modelMapping[machine] ?? machine
+    }
+
+    private static func chipName(for model: String) -> String {
+        let lower = model.lowercased()
+        if lower.contains("iphone 16 pro") || lower.contains("iphone 16 pro max") {
+            return "Apple A18 Pro"
+        } else if lower.contains("iphone 16") {
+            return "Apple A18"
+        } else if lower.contains("iphone 15 pro") || lower.contains("iphone 15 pro max") {
+            return "Apple A17 Pro"
+        } else if lower.contains("iphone 15") {
+            return "Apple A16 Bionic"
+        } else if lower.contains("iphone 14 pro") || lower.contains("iphone 14 pro max") {
+            return "Apple A16 Bionic"
+        } else if lower.contains("iphone 14") {
+            return "Apple A15 Bionic"
+        }
+        return "Apple Silicon"
+    }
+
+    private static func availableMemory() -> Int {
+        return Int(os_proc_available_memory())
+    }
+
+    private static let modelMapping: [String: String] = [
+        // iPhone 16 series
+        "iPhone17,1": "iPhone 16 Pro",
+        "iPhone17,2": "iPhone 16 Pro Max",
+        "iPhone17,3": "iPhone 16",
+        "iPhone17,4": "iPhone 16 Plus",
+        // iPhone 15 series
+        "iPhone16,1": "iPhone 15 Pro",
+        "iPhone16,2": "iPhone 15 Pro Max",
+        "iPhone15,4": "iPhone 15",
+        "iPhone15,5": "iPhone 15 Plus",
+        // iPhone 14 series
+        "iPhone15,2": "iPhone 14 Pro",
+        "iPhone15,3": "iPhone 14 Pro Max",
+        "iPhone14,7": "iPhone 14",
+        "iPhone14,8": "iPhone 14 Plus",
+        // iPhone 13 series
+        "iPhone14,2": "iPhone 13 Pro",
+        "iPhone14,3": "iPhone 13 Pro Max",
+        "iPhone14,5": "iPhone 13",
+        "iPhone14,4": "iPhone 13 mini",
+        // Simulator
+        "arm64": "iPhone (Simulator)",
+        "x86_64": "iPhone (Simulator)",
+    ]
+}
@@ -0,0 +1,123 @@
+import Foundation
+import Network
+import os
+
+struct DiscoveredCluster: Identifiable, Equatable {
+    let id: String
+    let name: String
+    let endpoint: NWEndpoint
+
+    static func == (lhs: DiscoveredCluster, rhs: DiscoveredCluster) -> Bool {
+        lhs.id == rhs.id && lhs.name == rhs.name
+    }
+}
+
+@Observable
+@MainActor
+final class DiscoveryService {
+    private(set) var discoveredClusters: [DiscoveredCluster] = []
+    private(set) var isSearching = false
+
+    private var browser: NWBrowser?
+
+    func startBrowsing() {
+        guard browser == nil else { return }
+
+        let browser = NWBrowser(for: .bonjour(type: "_exo._tcp", domain: nil), using: .tcp)
+
+        browser.stateUpdateHandler = { [weak self] state in
+            guard let service = self else { return }
+            Task { @MainActor in
+                switch state {
+                case .ready:
+                    service.isSearching = true
+                case .failed, .cancelled:
+                    service.isSearching = false
+                default:
+                    break
+                }
+            }
+        }
+
+        browser.browseResultsChangedHandler = { [weak self] results, _ in
+            guard let service = self else { return }
+            Task { @MainActor in
+                service.discoveredClusters = results.compactMap { result in
+                    guard case .service(let name, _, _, _) = result.endpoint else {
+                        return nil
+                    }
+                    return DiscoveredCluster(
+                        id: name,
+                        name: name,
+                        endpoint: result.endpoint
+                    )
+                }
+            }
+        }
+
+        browser.start(queue: .main)
+        self.browser = browser
+    }
+
+    func stopBrowsing() {
+        browser?.cancel()
+        browser = nil
+        isSearching = false
+        discoveredClusters = []
+    }
+
+    /// Resolve a discovered Bonjour endpoint to an IP address and port, then return a ConnectionInfo.
+    func resolve(_ cluster: DiscoveredCluster) async -> ConnectionInfo? {
+        await withCheckedContinuation { continuation in
+            let didResume = OSAllocatedUnfairLock(initialState: false)
+            let connection = NWConnection(to: cluster.endpoint, using: .tcp)
+            connection.stateUpdateHandler = { state in
+                guard
+                    didResume.withLock({
+                        guard !$0 else { return false }
+                        $0 = true
+                        return true
+                    })
+                else { return }
+                switch state {
+                case .ready:
+                    if let innerEndpoint = connection.currentPath?.remoteEndpoint,
+                        case .hostPort(let host, let port) = innerEndpoint
+                    {
+                        var hostString: String
+                        switch host {
+                        case .ipv4(let addr):
+                            hostString = "\(addr)"
+                        case .ipv6(let addr):
+                            hostString = "\(addr)"
+                        case .name(let name, _):
+                            hostString = name
+                        @unknown default:
+                            hostString = "\(host)"
+                        }
+                        // Strip interface scope suffix (e.g. "%en0")
+                        if let pct = hostString.firstIndex(of: "%") {
+                            hostString = String(hostString[..<pct])
+                        }
+                        let info = ConnectionInfo(
+                            host: hostString,
+                            port: Int(port.rawValue),
+                            nodeId: nil
+                        )
+                        connection.cancel()
+                        continuation.resume(returning: info)
+                    } else {
+                        connection.cancel()
+                        continuation.resume(returning: nil)
+                    }
+                case .failed, .cancelled:
+                    continuation.resume(returning: nil)
+                default:
+                    // Not a terminal state — allow future callbacks
+                    didResume.withLock { $0 = false }
+                }
+            }
+            connection.start(queue: .global(qos: .userInitiated))
+        }
+    }
+}
@@ -0,0 +1,201 @@
+import Foundation
+import MLXLLM
+import MLXLMCommon
+
+enum LocalModelState: Equatable {
+    case notDownloaded
+    case downloading(progress: Double)
+    case downloaded
+    case loading
+    case ready
+    case generating
+    case error(String)
+}
+
+@Observable
+@MainActor
+final class LocalInferenceService {
+    private(set) var modelState: LocalModelState = .notDownloaded
+    private var modelContainer: ModelContainer?
+    private var generationTask: Task<Void, Never>?
+
+    let defaultModelId = "mlx-community/Qwen3-0.6B-4bit"
+
+    private static let modelDownloadedKey = "exo_local_model_downloaded"
+
+    var isReady: Bool {
+        modelState == .ready
+    }
+
+    var isAvailable: Bool {
+        modelState == .ready || modelState == .generating
+    }
+
+    init() {
+        if UserDefaults.standard.bool(forKey: Self.modelDownloadedKey) {
+            modelState = .downloaded
+        }
+    }
+
+    // MARK: - Model Lifecycle
+
+    func prepareModel() async {
+        guard modelState == .notDownloaded || modelState == .downloaded else { return }
+
+        let wasDownloaded = modelState == .downloaded
+
+        if !wasDownloaded {
+            modelState = .downloading(progress: 0)
+        } else {
+            modelState = .loading
+        }
+
+        do {
+            let container = try await loadModelContainer(
+                id: defaultModelId
+            ) { [weak self] progress in
+                guard let self else { return }
+                Task { @MainActor in
+                    if case .downloading = self.modelState {
+                        self.modelState = .downloading(progress: progress.fractionCompleted)
+                    }
+                }
+            }
+
+            self.modelContainer = container
+            UserDefaults.standard.set(true, forKey: Self.modelDownloadedKey)
+            modelState = .ready
+        } catch {
+            modelState = .error(error.localizedDescription)
+        }
+    }
+
+    func unloadModel() {
+        cancelGeneration()
+        modelContainer = nil
+        modelState = .downloaded
+    }
+
+    // MARK: - Generation
+
+    func streamChatCompletion(request: ChatCompletionRequest) -> AsyncThrowingStream<
+        ChatCompletionChunk, Error
+    > {
+        AsyncThrowingStream { continuation in
+            let task = Task { [weak self] in
+                guard let self else {
+                    continuation.finish(throwing: LocalInferenceError.serviceUnavailable)
+                    return
+                }
+
+                guard let container = self.modelContainer else {
+                    continuation.finish(throwing: LocalInferenceError.modelNotLoaded)
+                    return
+                }
+
+                await MainActor.run {
+                    self.modelState = .generating
+                }
+
+                defer {
+                    Task { @MainActor [weak self] in
+                        if self?.modelState == .generating {
+                            self?.modelState = .ready
+                        }
+                    }
+                }
+
+                let chunkId = "local-\(UUID().uuidString)"
+
+                do {
+                    // Build Chat.Message array from the request
+                    var chatMessages: [Chat.Message] = []
+                    for msg in request.messages {
+                        switch msg.role {
+                        case "system":
+                            chatMessages.append(.system(msg.content))
+                        case "assistant":
+                            chatMessages.append(.assistant(msg.content))
+                        default:
+                            chatMessages.append(.user(msg.content))
+                        }
+                    }
+
+                    // Use ChatSession for streaming generation
+                    let session = ChatSession(
+                        container,
+                        history: chatMessages,
+                        generateParameters: GenerateParameters(
+                            maxTokens: request.maxTokens ?? 4096,
+                            temperature: Float(request.temperature ?? 0.7)
+                        )
+                    )
+
+                    // Stream with an empty prompt since history already contains the conversation
+                    let stream = session.streamResponse(to: "")
+                    for try await text in stream {
+                        if Task.isCancelled { break }
+
+                        let chunk = ChatCompletionChunk(
+                            id: chunkId,
+                            model: request.model,
+                            choices: [
+                                StreamingChoice(
+                                    index: 0,
+                                    delta: Delta(role: nil, content: text),
+                                    finishReason: nil
+                                )
+                            ],
+                            usage: nil
+                        )
+                        continuation.yield(chunk)
+                    }
+
+                    // Send final chunk with finish reason
+                    let finalChunk = ChatCompletionChunk(
+                        id: chunkId,
+                        model: request.model,
+                        choices: [
+                            StreamingChoice(
+                                index: 0,
+                                delta: Delta(role: nil, content: nil),
+                                finishReason: "stop"
+                            )
+                        ],
+                        usage: nil
+                    )
+                    continuation.yield(finalChunk)
+                    continuation.finish()
+                } catch {
+                    continuation.finish(throwing: error)
+                }
+            }
+
+            self.generationTask = task
+
+            continuation.onTermination = { _ in
+                task.cancel()
+            }
+        }
+    }
+
+    func cancelGeneration() {
+        generationTask?.cancel()
+        generationTask = nil
+        if modelState == .generating {
+            modelState = .ready
+        }
+    }
+}
+
+enum LocalInferenceError: LocalizedError {
+    case serviceUnavailable
+    case modelNotLoaded
+
+    var errorDescription: String? {
+        switch self {
+        case .serviceUnavailable: "Local inference service is unavailable"
+        case .modelNotLoaded: "Local model is not loaded"
+        }
+    }
+}
@@ -0,0 +1,50 @@
+import Foundation
+
+struct SSEStreamParser<T: Decodable>: AsyncSequence {
+    typealias Element = T
+
+    let bytes: URLSession.AsyncBytes
+    let decoder: JSONDecoder
+
+    init(bytes: URLSession.AsyncBytes, decoder: JSONDecoder = JSONDecoder()) {
+        self.bytes = bytes
+        self.decoder = decoder
+    }
+
+    func makeAsyncIterator() -> AsyncIterator {
+        AsyncIterator(lines: bytes.lines, decoder: decoder)
+    }
+
+    struct AsyncIterator: AsyncIteratorProtocol {
+        var lines: AsyncLineSequence<URLSession.AsyncBytes>.AsyncIterator
+        let decoder: JSONDecoder
+
+        init(lines: AsyncLineSequence<URLSession.AsyncBytes>, decoder: JSONDecoder) {
+            self.lines = lines.makeAsyncIterator()
+            self.decoder = decoder
+        }
+
+        mutating func next() async throws -> T? {
+            while let line = try await lines.next() {
+                let trimmed = line.trimmingCharacters(in: .whitespaces)
+
+                guard trimmed.hasPrefix("data: ") else { continue }
+
+                let payload = String(trimmed.dropFirst(6))
+
+                if payload == "[DONE]" {
+                    return nil
+                }
+
+                guard let data = payload.data(using: .utf8) else { continue }
+
+                do {
+                    return try decoder.decode(T.self, from: data)
+                } catch {
+                    continue
+                }
+            }
+            return nil
+        }
+    }
+}
@@ -0,0 +1,203 @@
+import SwiftUI
+
+struct ChatView: View {
+    @Environment(ClusterService.self) private var clusterService
+    @Environment(ChatService.self) private var chatService
+    @Environment(LocalInferenceService.self) private var localInferenceService
+    @State private var inputText = ""
+    @State private var showModelSelector = false
+
+    var body: some View {
+        VStack(spacing: 0) {
+            modelBar
+
+            GradientDivider()
+
+            messageList
+
+            GradientDivider()
+
+            inputBar
+        }
+        .background(Color.exoBlack)
+        .sheet(isPresented: $showModelSelector) {
+            ModelSelectorView(
+                models: clusterService.availableModels,
+                selectedModelId: chatService.activeConversation?.modelId
+            ) { modelId in
+                chatService.setModelForActiveConversation(modelId)
+            }
+            .presentationBackground(Color.exoDarkGray)
+        }
+    }
+
+    // MARK: - Model Bar
+
+    private var useLocalModel: Bool {
+        !clusterService.isConnected && localInferenceService.isAvailable
+    }
+
+    private var modelBar: some View {
+        Button {
+            if !useLocalModel {
+                showModelSelector = true
+            }
+        } label: {
+            HStack {
+                Image(systemName: useLocalModel ? "iphone" : "cpu")
+                    .font(.exoCaption)
+                    .foregroundStyle(useLocalModel ? Color.exoYellow : Color.exoLightGray)
+
+                if useLocalModel {
+                    Text(localInferenceService.defaultModelId)
+                        .font(.exoSubheadline)
+                        .foregroundStyle(Color.exoForeground)
+                        .lineLimit(1)
+                } else if let modelId = chatService.activeConversation?.modelId {
+                    Text(modelId)
+                        .font(.exoSubheadline)
+                        .foregroundStyle(Color.exoForeground)
+                        .lineLimit(1)
+                } else {
+                    Text("SELECT MODEL")
+                        .font(.exoSubheadline)
+                        .tracking(1.5)
+                        .foregroundStyle(Color.exoLightGray)
+                }
+
+                Spacer()
+
+                if useLocalModel {
+                    Text("ON-DEVICE")
+                        .font(.exoCaption)
+                        .tracking(1)
+                        .foregroundStyle(Color.exoYellow)
+                        .padding(.horizontal, 6)
+                        .padding(.vertical, 2)
+                        .background(Color.exoYellow.opacity(0.15))
+                        .clipShape(Capsule())
+                } else {
+                    Image(systemName: "chevron.right")
+                        .font(.caption)
+                        .foregroundStyle(Color.exoLightGray)
+                }
+            }
+            .padding(.horizontal)
+            .padding(.vertical, 10)
+            .background(Color.exoDarkGray)
+        }
+        .tint(.primary)
+        .disabled(useLocalModel)
+    }
+
+    // MARK: - Messages
+
+    private var messageList: some View {
+        ScrollViewReader { proxy in
+            ScrollView {
+                LazyVStack(spacing: 12) {
+                    if chatService.activeMessages.isEmpty {
+                        emptyState
+                    } else {
+                        ForEach(chatService.activeMessages) { message in
+                            MessageBubbleView(message: message)
+                                .id(message.id)
+                        }
+                    }
+                }
+                .padding()
+            }
+            .background(Color.exoBlack)
+            .onChange(of: chatService.activeMessages.last?.content) {
+                if let lastId = chatService.activeMessages.last?.id {
+                    withAnimation(.easeOut(duration: 0.2)) {
+                        proxy.scrollTo(lastId, anchor: .bottom)
+                    }
+                }
+            }
+        }
+    }
+
+    private var emptyState: some View {
+        VStack(spacing: 16) {
+            Spacer(minLength: 80)
+
+            ZStack {
+                Circle()
+                    .stroke(Color.exoYellow.opacity(0.15), lineWidth: 1)
+                    .frame(width: 80, height: 80)
+                Circle()
+                    .stroke(Color.exoYellow.opacity(0.3), lineWidth: 1)
+                    .frame(width: 56, height: 56)
+                Circle()
+                    .fill(Color.exoYellow.opacity(0.15))
+                    .frame(width: 32, height: 32)
+                Circle()
+                    .fill(Color.exoYellow)
+                    .frame(width: 8, height: 8)
+                    .shadow(color: Color.exoYellow.opacity(0.6), radius: 6)
+            }
+
+            Text("AWAITING INPUT")
+                .font(.exoSubheadline)
+                .tracking(3)
+                .foregroundStyle(Color.exoLightGray)
+
+            Text("Send a message to begin.")
+                .font(.exoCaption)
+                .foregroundStyle(Color.exoLightGray.opacity(0.6))
+
+            Spacer(minLength: 80)
+        }
+        .padding()
+    }
+
+    // MARK: - Input
+
+    private var inputBar: some View {
+        HStack(alignment: .bottom, spacing: 8) {
+            TextField("Message...", text: $inputText, axis: .vertical)
+                .font(.exoBody)
+                .lineLimit(1...6)
+                .textFieldStyle(.plain)
+                .padding(10)
+                .background(Color.exoMediumGray)
+                .foregroundStyle(Color.exoForeground)
+                .clipShape(RoundedRectangle(cornerRadius: 8))
+
+            if chatService.isGenerating {
+                Button {
+                    chatService.cancelGeneration()
+                } label: {
+                    Image(systemName: "stop.circle.fill")
+                        .font(.title2)
+                        .foregroundStyle(Color.exoDestructive)
+                }
+            } else {
+                Button {
+                    let text = inputText
+                    inputText = ""
+                    chatService.sendMessage(text)
+                } label: {
+                    Text("SEND")
+                        .font(.exoMono(12, weight: .bold))
+                        .tracking(1)
+                        .foregroundStyle(canSend ? Color.exoBlack : Color.exoLightGray)
+                        .padding(.horizontal, 14)
+                        .padding(.vertical, 8)
+                        .background(canSend ? Color.exoYellow : Color.exoMediumGray)
+                        .clipShape(RoundedRectangle(cornerRadius: 8))
+                }
+                .disabled(!canSend)
+            }
+        }
+        .padding(.horizontal)
+        .padding(.vertical, 8)
+        .background(Color.exoDarkGray)
+    }
+
+    private var canSend: Bool {
+        !inputText.trimmingCharacters(in: .whitespacesAndNewlines).isEmpty
+            && (clusterService.isConnected || localInferenceService.isAvailable)
+    }
+}
@@ -0,0 +1,54 @@
+import SwiftUI
+
+struct MessageBubbleView: View {
+    let message: ChatMessage
+
+    private var isAssistant: Bool { message.role == .assistant }
+
+    var body: some View {
+        HStack {
+            if message.role == .user { Spacer(minLength: 48) }
+
+            VStack(alignment: isAssistant ? .leading : .trailing, spacing: 6) {
+                // Header
+                HStack(spacing: 4) {
+                    if isAssistant {
+                        Circle()
+                            .fill(Color.exoYellow)
+                            .frame(width: 6, height: 6)
+                            .shadow(color: Color.exoYellow.opacity(0.6), radius: 4)
+                        Text("EXO")
+                            .font(.exoMono(10, weight: .bold))
+                            .tracking(1.5)
+                            .foregroundStyle(Color.exoYellow)
+                    } else {
+                        Text("QUERY")
+                            .font(.exoMono(10, weight: .medium))
+                            .tracking(1.5)
+                            .foregroundStyle(Color.exoLightGray)
+                    }
+                }
+
+                // Bubble
+                HStack(spacing: 0) {
+                    if isAssistant {
+                        RoundedRectangle(cornerRadius: 1)
+                            .fill(Color.exoYellow.opacity(0.5))
+                            .frame(width: 2)
+                    }
+
+                    Text(message.content + (message.isStreaming ? " \u{258C}" : ""))
+                        .font(.exoBody)
+                        .textSelection(.enabled)
+                        .foregroundStyle(Color.exoForeground)
+                        .padding(.horizontal, 14)
+                        .padding(.vertical, 10)
+                }
+                .background(Color.exoDarkGray)
+                .clipShape(RoundedRectangle(cornerRadius: 8))
+            }
+
+            if isAssistant { Spacer(minLength: 48) }
+        }
+    }
+}
@@ -0,0 +1,75 @@
+import SwiftUI
+
+struct ModelSelectorView: View {
+    let models: [ModelOption]
+    let selectedModelId: String?
+    let onSelect: (String) -> Void
+    @Environment(\.dismiss) private var dismiss
+
+    var body: some View {
+        NavigationStack {
+            List {
+                if models.isEmpty {
+                    emptyContent
+                } else {
+                    modelsList
+                }
+            }
+            .scrollContentBackground(.hidden)
+            .background(Color.exoBlack)
+            .navigationTitle("SELECT MODEL")
+            .navigationBarTitleDisplayMode(.inline)
+            .toolbar {
+                ToolbarItem(placement: .cancellationAction) {
+                    Button("Cancel") { dismiss() }
+                        .font(.exoSubheadline)
+                        .foregroundStyle(Color.exoYellow)
+                }
+            }
+        }
+    }
+
+    private var emptyContent: some View {
+        ContentUnavailableView(
+            "No Models Available",
+            systemImage: "cpu",
+            description: Text("Connect to an EXO cluster to see available models.")
+        )
+        .foregroundStyle(Color.exoLightGray)
+        .listRowBackground(Color.exoBlack)
+    }
+
+    private var modelsList: some View {
+        ForEach(models) { model in
+            Button {
+                onSelect(model.id)
+                dismiss()
+            } label: {
+                modelRow(model)
+            }
+            .tint(.primary)
+            .listRowBackground(Color.exoDarkGray)
+        }
+    }
+
+    private func modelRow(_ model: ModelOption) -> some View {
+        HStack {
+            VStack(alignment: .leading, spacing: 2) {
+                Text(model.displayName)
+                    .font(.exoSubheadline)
+                    .foregroundStyle(Color.exoForeground)
+                Text(model.id)
+                    .font(.exoCaption)
+                    .foregroundStyle(Color.exoLightGray)
+            }
+
+            Spacer()
+
+            if model.id == selectedModelId {
+                Image(systemName: "checkmark")
+                    .font(.exoSubheadline)
+                    .foregroundStyle(Color.exoYellow)
+            }
+        }
+    }
+}
@@ -0,0 +1,68 @@
+import SwiftUI
+
+struct ConnectionStatusBadge: View {
+    let connectionState: ConnectionState
+    var localModelState: LocalModelState = .notDownloaded
+
+    private var isLocalReady: Bool {
+        if case .disconnected = connectionState {
+            return localModelState == .ready || localModelState == .generating
+        }
+        return false
+    }
+
+    var body: some View {
+        HStack(spacing: 6) {
+            Circle()
+                .fill(dotColor)
+                .frame(width: 8, height: 8)
+                .shadow(color: dotColor.opacity(0.6), radius: 4)
+
+            Text(label.uppercased())
+                .font(.exoMono(10, weight: .medium))
+                .tracking(1)
+                .foregroundStyle(Color.exoForeground)
+        }
+        .padding(.horizontal, 10)
+        .padding(.vertical, 5)
+        .background(backgroundColor)
+        .clipShape(Capsule())
+        .overlay(
+            Capsule()
+                .stroke(dotColor.opacity(0.3), lineWidth: 1)
+        )
+    }
+
+    private var dotColor: Color {
+        if isLocalReady {
+            return .exoYellow
+        }
+        switch connectionState {
+        case .connected: return .green
+        case .connecting: return .orange
+        case .disconnected: return .exoLightGray
+        }
+    }
+
+    private var label: String {
+        if isLocalReady {
+            return "Local"
+        }
+        switch connectionState {
+        case .connected: return "Connected"
+        case .connecting: return "Connecting"
+        case .disconnected: return "Disconnected"
+        }
+    }
+
+    private var backgroundColor: Color {
+        if isLocalReady {
+            return Color.exoYellow.opacity(0.1)
+        }
+        switch connectionState {
+        case .connected: return .green.opacity(0.1)
+        case .connecting: return .orange.opacity(0.1)
+        case .disconnected: return Color.exoMediumGray.opacity(0.5)
+        }
+    }
+}
@@ -0,0 +1,136 @@
+import SwiftUI
+
+struct RootView: View {
+    @Environment(ClusterService.self) private var clusterService
+    @Environment(DiscoveryService.self) private var discoveryService
+    @Environment(ChatService.self) private var chatService
+    @Environment(LocalInferenceService.self) private var localInferenceService
+    @State private var showSettings = false
+    @State private var showConversations = false
+
+    var body: some View {
+        NavigationStack {
+            ChatView()
+                .navigationTitle("EXO")
+                .navigationBarTitleDisplayMode(.inline)
+                .toolbar {
+                    ToolbarItem(placement: .topBarLeading) {
+                        conversationMenuButton
+                    }
+
+                    ToolbarItem(placement: .principal) {
+                        ConnectionStatusBadge(
+                            connectionState: clusterService.connectionState,
+                            localModelState: localInferenceService.modelState
+                        )
+                    }
+
+                    ToolbarItem(placement: .topBarTrailing) {
+                        Button {
+                            showSettings = true
+                        } label: {
+                            Image(systemName: "gear")
+                                .foregroundStyle(Color.exoYellow)
+                        }
+                    }
+                }
+        }
+        .tint(Color.exoYellow)
+        .sheet(isPresented: $showSettings) {
+            SettingsView()
+                .environment(discoveryService)
+                .presentationBackground(Color.exoDarkGray)
+        }
+        .sheet(isPresented: $showConversations) {
+            conversationList
+                .presentationBackground(Color.exoDarkGray)
+        }
+    }
+
+    // MARK: - Conversations
+
+    private var conversationMenuButton: some View {
+        HStack(spacing: 12) {
+            Button {
+                showConversations = true
+            } label: {
+                Image(systemName: "sidebar.left")
+                    .foregroundStyle(Color.exoYellow)
+            }
+
+            Button {
+                chatService.createConversation()
+            } label: {
+                Image(systemName: "square.and.pencil")
+                    .foregroundStyle(Color.exoYellow)
+            }
+        }
+    }
+
+    private var conversationList: some View {
+        NavigationStack {
+            List {
+                if chatService.conversations.isEmpty {
+                    Text("No conversations yet")
+                        .font(.exoBody)
+                        .foregroundStyle(Color.exoLightGray)
+                        .listRowBackground(Color.exoDarkGray)
+                } else {
+                    ForEach(chatService.conversations) { conversation in
+                        let isActive = conversation.id == chatService.activeConversationId
+                        Button {
+                            chatService.setActiveConversation(id: conversation.id)
+                            showConversations = false
+                        } label: {
+                            VStack(alignment: .leading, spacing: 4) {
+                                Text(conversation.title)
+                                    .font(.exoSubheadline)
+                                    .fontWeight(isActive ? .semibold : .regular)
+                                    .foregroundStyle(
+                                        isActive ? Color.exoYellow : Color.exoForeground
+                                    )
+                                    .lineLimit(1)
+
+                                if let modelId = conversation.modelId {
+                                    Text(modelId)
+                                        .font(.exoCaption)
+                                        .foregroundStyle(Color.exoLightGray)
+                                        .lineLimit(1)
+                                }
+                            }
+                        }
+                        .listRowBackground(
+                            isActive
+                                ? Color.exoYellow.opacity(0.1)
+                                : Color.exoDarkGray
+                        )
+                    }
+                    .onDelete { indexSet in
+                        for index in indexSet {
+                            chatService.deleteConversation(id: chatService.conversations[index].id)
+                        }
+                    }
+                }
+            }
+            .scrollContentBackground(.hidden)
+            .background(Color.exoBlack)
+            .navigationTitle("Conversations")
+            .navigationBarTitleDisplayMode(.inline)
+            .toolbar {
+                ToolbarItem(placement: .confirmationAction) {
+                    Button("Done") { showConversations = false }
+                        .font(.exoSubheadline)
+                        .foregroundStyle(Color.exoYellow)
+                }
+                ToolbarItem(placement: .topBarLeading) {
+                    Button {
+                        chatService.createConversation()
+                    } label: {
+                        Image(systemName: "plus")
+                            .foregroundStyle(Color.exoYellow)
+                    }
+                }
+            }
+        }
+    }
+}
@@ -0,0 +1,314 @@
+import SwiftUI
+
+struct SettingsView: View {
+    @Environment(ClusterService.self) private var clusterService
+    @Environment(DiscoveryService.self) private var discoveryService
+    @Environment(LocalInferenceService.self) private var localInferenceService
+    @Environment(\.dismiss) private var dismiss
+    @State private var host: String = ""
+    @State private var port: String = "52415"
+
+    var body: some View {
+        NavigationStack {
+            Form {
+                localModelSection
+                nearbyClustersSection
+                connectionSection
+                statusSection
+            }
+            .scrollContentBackground(.hidden)
+            .background(Color.exoBlack)
+            .navigationTitle("Settings")
+            .navigationBarTitleDisplayMode(.inline)
+            .toolbar {
+                ToolbarItem(placement: .confirmationAction) {
+                    Button("Done") { dismiss() }
+                        .font(.exoSubheadline)
+                        .foregroundStyle(Color.exoYellow)
+                }
+            }
+        }
+    }
+
+    // MARK: - Section Headers
+
+    private func sectionHeader(_ title: String) -> some View {
+        Text(title.uppercased())
+            .font(.exoMono(10, weight: .semibold))
+            .tracking(2)
+            .foregroundStyle(Color.exoYellow)
+    }
+
+    // MARK: - Local Model
+
+    private var localModelSection: some View {
+        Section {
+            HStack {
+                VStack(alignment: .leading, spacing: 4) {
+                    Text(localInferenceService.defaultModelId)
+                        .font(.exoSubheadline)
+                        .foregroundStyle(Color.exoForeground)
+
+                    Text(localModelStatusText)
+                        .font(.exoCaption)
+                        .foregroundStyle(Color.exoLightGray)
+                }
+
+                Spacer()
+
+                localModelActionButton
+            }
+            .listRowBackground(Color.exoDarkGray)
+
+            if case .downloading(let progress) = localInferenceService.modelState {
+                ProgressView(value: progress)
+                    .tint(Color.exoYellow)
+                    .listRowBackground(Color.exoDarkGray)
+            }
+        } header: {
+            sectionHeader("Local Model")
+        } footer: {
+            Text(
+                "When disconnected from a cluster, messages are processed on-device using this model."
+            )
+            .font(.exoCaption)
+            .foregroundStyle(Color.exoLightGray.opacity(0.7))
+        }
+    }
+
+    private var localModelStatusText: String {
+        switch localInferenceService.modelState {
+        case .notDownloaded: "Not downloaded"
+        case .downloading(let progress): "Downloading \(Int(progress * 100))%..."
+        case .downloaded: "Downloaded — not loaded"
+        case .loading: "Loading into memory..."
+        case .ready: "Ready"
+        case .generating: "Generating..."
+        case .error(let message): "Error: \(message)"
+        }
+    }
+
+    @ViewBuilder
+    private var localModelActionButton: some View {
+        switch localInferenceService.modelState {
+        case .notDownloaded:
+            exoButton("Download") {
+                Task { await localInferenceService.prepareModel() }
+            }
+        case .downloading:
+            ProgressView()
+                .controlSize(.small)
+                .tint(Color.exoYellow)
+        case .downloaded:
+            exoButton("Load") {
+                Task { await localInferenceService.prepareModel() }
+            }
+        case .loading:
+            ProgressView()
+                .controlSize(.small)
+                .tint(Color.exoYellow)
+        case .ready, .generating:
+            exoButton("Unload") {
+                localInferenceService.unloadModel()
+            }
+        case .error:
+            exoButton("Retry", destructive: true) {
+                Task { await localInferenceService.prepareModel() }
+            }
+        }
+    }
+
+    private func exoButton(_ title: String, destructive: Bool = false, action: @escaping () -> Void)
+        -> some View
+    {
+        let borderColor = destructive ? Color.exoDestructive : Color.exoYellow
+        return Button(action: action) {
+            Text(title.uppercased())
+                .font(.exoMono(11, weight: .semibold))
+                .tracking(1)
+                .foregroundStyle(borderColor)
+                .padding(.horizontal, 10)
+                .padding(.vertical, 5)
+                .overlay(
+                    RoundedRectangle(cornerRadius: 6)
+                        .stroke(borderColor, lineWidth: 1)
+                )
+        }
+    }
+
+    // MARK: - Nearby Clusters
+
+    private var nearbyClustersSection: some View {
+        Section {
+            if discoveryService.discoveredClusters.isEmpty {
+                if discoveryService.isSearching {
+                    HStack {
+                        ProgressView()
+                            .tint(Color.exoYellow)
+                            .padding(.trailing, 8)
+                        Text("Searching for clusters...")
+                            .font(.exoBody)
+                            .foregroundStyle(Color.exoLightGray)
+                    }
+                    .listRowBackground(Color.exoDarkGray)
+                } else {
+                    Text("No clusters found")
+                        .font(.exoBody)
+                        .foregroundStyle(Color.exoLightGray)
+                        .listRowBackground(Color.exoDarkGray)
+                }
+            } else {
+                ForEach(discoveryService.discoveredClusters) { cluster in
+                    HStack {
+                        VStack(alignment: .leading) {
+                            Text(cluster.name)
+                                .font(.exoBody)
+                                .foregroundStyle(Color.exoForeground)
+                        }
+                        Spacer()
+                        exoButton("Connect") {
+                            Task {
+                                await clusterService.connectToDiscoveredCluster(
+                                    cluster, using: discoveryService
+                                )
+                                if clusterService.isConnected {
+                                    dismiss()
+                                }
+                            }
+                        }
+                    }
+                    .listRowBackground(Color.exoDarkGray)
+                }
+            }
+        } header: {
+            sectionHeader("Nearby Clusters")
+        }
+    }
+
+    // MARK: - Manual Connection
+
+    private var connectionSection: some View {
+        Section {
+            TextField("IP Address (e.g. 192.168.1.42)", text: $host)
+                .font(.exoBody)
+                .keyboardType(.decimalPad)
+                .textContentType(.URL)
+                .autocorrectionDisabled()
+                .foregroundStyle(Color.exoForeground)
+                .listRowBackground(Color.exoDarkGray)
+
+            TextField("Port", text: $port)
+                .font(.exoBody)
+                .keyboardType(.numberPad)
+                .foregroundStyle(Color.exoForeground)
+                .listRowBackground(Color.exoDarkGray)
+
+            Button {
+                Task {
+                    let portNum = Int(port) ?? ConnectionInfo.defaultPort
+                    let info = ConnectionInfo(host: host, port: portNum, nodeId: nil)
+                    await clusterService.connect(to: info)
+                    if clusterService.isConnected {
+                        dismiss()
+                    }
+                }
+            } label: {
+                Text(clusterService.isConnected ? "RECONNECT" : "CONNECT")
+                    .font(.exoMono(13, weight: .semibold))
+                    .tracking(1.5)
+                    .foregroundStyle(
+                        host.trimmingCharacters(in: .whitespacesAndNewlines).isEmpty
+                            ? Color.exoLightGray : Color.exoYellow
+                    )
+                    .frame(maxWidth: .infinity)
+            }
+            .disabled(host.trimmingCharacters(in: .whitespacesAndNewlines).isEmpty)
+            .listRowBackground(Color.exoDarkGray)
+        } header: {
+            sectionHeader("Manual Connection")
+        }
+    }
+
+    // MARK: - Status
+
+    private var statusSection: some View {
+        Section {
+            if let connection = clusterService.currentConnection {
+                LabeledContent {
+                    Text(connection.host)
+                        .font(.exoCaption)
+                        .foregroundStyle(Color.exoForeground)
+                } label: {
+                    Text("Host")
+                        .font(.exoCaption)
+                        .foregroundStyle(Color.exoLightGray)
+                }
+                .listRowBackground(Color.exoDarkGray)
+
+                LabeledContent {
+                    Text("\(connection.port)")
+                        .font(.exoCaption)
+                        .foregroundStyle(Color.exoForeground)
+                } label: {
+                    Text("Port")
+                        .font(.exoCaption)
+                        .foregroundStyle(Color.exoLightGray)
+                }
+                .listRowBackground(Color.exoDarkGray)
+
+                if let nodeId = connection.nodeId {
+                    LabeledContent {
+                        Text(String(nodeId.prefix(12)) + "...")
+                            .font(.exoCaption)
+                            .foregroundStyle(Color.exoForeground)
+                    } label: {
+                        Text("Node ID")
+                            .font(.exoCaption)
+                            .foregroundStyle(Color.exoLightGray)
+                    }
+                    .listRowBackground(Color.exoDarkGray)
+                }
+
+                LabeledContent {
+                    Text("\(clusterService.availableModels.count)")
+                        .font(.exoCaption)
+                        .foregroundStyle(Color.exoForeground)
+                } label: {
+                    Text("Models")
+                        .font(.exoCaption)
+                        .foregroundStyle(Color.exoLightGray)
+                }
+                .listRowBackground(Color.exoDarkGray)
+
+                Button(role: .destructive) {
+                    clusterService.disconnect()
+                } label: {
+                    Text("DISCONNECT")
+                        .font(.exoMono(13, weight: .semibold))
+                        .tracking(1.5)
+                        .foregroundStyle(Color.exoDestructive)
+                        .frame(maxWidth: .infinity)
+                }
+                .listRowBackground(Color.exoDarkGray)
+            } else {
+                if let error = clusterService.lastError {
+                    Label {
+                        Text(error)
+                            .font(.exoCaption)
+                    } icon: {
+                        Image(systemName: "exclamationmark.triangle")
+                    }
+                    .foregroundStyle(Color.exoDestructive)
+                    .listRowBackground(Color.exoDarkGray)
+                } else {
+                    Text("Not connected")
+                        .font(.exoBody)
+                        .foregroundStyle(Color.exoLightGray)
+                        .listRowBackground(Color.exoDarkGray)
+                }
+            }
+        } header: {
+            sectionHeader("Status")
+        }
+    }
+}
@@ -0,0 +1,51 @@
+import SwiftUI
+
+// MARK: - EXO Color Palette
+
+extension Color {
+    /// Primary background — near-black (#121212)
+    static let exoBlack = Color(red: 0x12 / 255.0, green: 0x12 / 255.0, blue: 0x12 / 255.0)
+    /// Card / surface background (#1F1F1F)
+    static let exoDarkGray = Color(red: 0x1F / 255.0, green: 0x1F / 255.0, blue: 0x1F / 255.0)
+    /// Input field / elevated surface (#353535)
+    static let exoMediumGray = Color(red: 0x35 / 255.0, green: 0x35 / 255.0, blue: 0x35 / 255.0)
+    /// Secondary text (#999999)
+    static let exoLightGray = Color(red: 0x99 / 255.0, green: 0x99 / 255.0, blue: 0x99 / 255.0)
+    /// Accent yellow — matches dashboard (#FFD700)
+    static let exoYellow = Color(red: 0xFF / 255.0, green: 0xD7 / 255.0, blue: 0x00 / 255.0)
+    /// Primary foreground text (#E5E5E5)
+    static let exoForeground = Color(red: 0xE5 / 255.0, green: 0xE5 / 255.0, blue: 0xE5 / 255.0)
+    /// Destructive / error (#E74C3C)
+    static let exoDestructive = Color(red: 0xE7 / 255.0, green: 0x4C / 255.0, blue: 0x3C / 255.0)
+}
+
+// MARK: - EXO Typography (SF Mono via .monospaced design)
+
+extension Font {
+    /// Monospaced font at a given size and weight.
+    static func exoMono(_ size: CGFloat, weight: Font.Weight = .regular) -> Font {
+        .system(size: size, weight: weight, design: .monospaced)
+    }
+
+    /// Body text — 15pt monospaced
+    static let exoBody: Font = .system(size: 15, weight: .regular, design: .monospaced)
+    /// Caption — 11pt monospaced
+    static let exoCaption: Font = .system(size: 11, weight: .regular, design: .monospaced)
+    /// Subheadline — 13pt monospaced medium
+    static let exoSubheadline: Font = .system(size: 13, weight: .medium, design: .monospaced)
+    /// Headline — 17pt monospaced semibold
+    static let exoHeadline: Font = .system(size: 17, weight: .semibold, design: .monospaced)
+}
+
+// MARK: - Reusable Gradient Divider
+
+struct GradientDivider: View {
+    var body: some View {
+        LinearGradient(
+            colors: [.clear, Color.exoYellow.opacity(0.3), .clear],
+            startPoint: .leading,
+            endPoint: .trailing
+        )
+        .frame(height: 1)
+    }
+}
@@ -0,0 +1,18 @@
+//
+//  EXO_iOSTests.swift
+//  EXO-iOSTests
+//
+//  Created by Sami Khan on 2026-02-17.
+//
+
+import Testing
+
+@testable import EXO_iOS
+
+struct EXO_iOSTests {
+
+    @Test func example() async throws {
+        // Write your test here and use APIs like `#expect(...)` to check expected conditions.
+    }
+
+}
@@ -0,0 +1,41 @@
+//
+//  EXO_iOSUITests.swift
+//  EXO-iOSUITests
+//
+//  Created by Sami Khan on 2026-02-17.
+//
+
+import XCTest
+
+final class EXO_iOSUITests: XCTestCase {
+
+    override func setUpWithError() throws {
+        // Put setup code here. This method is called before the invocation of each test method in the class.
+
+        // In UI tests it is usually best to stop immediately when a failure occurs.
+        continueAfterFailure = false
+
+        // In UI tests it’s important to set the initial state - such as interface orientation - required for your tests before they run. The setUp method is a good place to do this.
+    }
+
+    override func tearDownWithError() throws {
+        // Put teardown code here. This method is called after the invocation of each test method in the class.
+    }
+
+    @MainActor
+    func testExample() throws {
+        // UI tests must launch the application that they test.
+        let app = XCUIApplication()
+        app.launch()
+
+        // Use XCTAssert and related functions to verify your tests produce the correct results.
+    }
+
+    @MainActor
+    func testLaunchPerformance() throws {
+        // This measures how long it takes to launch your application.
+        measure(metrics: [XCTApplicationLaunchMetric()]) {
+            XCUIApplication().launch()
+        }
+    }
+}
@@ -0,0 +1,33 @@
+//
+//  EXO_iOSUITestsLaunchTests.swift
+//  EXO-iOSUITests
+//
+//  Created by Sami Khan on 2026-02-17.
+//
+
+import XCTest
+
+final class EXO_iOSUITestsLaunchTests: XCTestCase {
+
+    override class var runsForEachTargetApplicationUIConfiguration: Bool {
+        true
+    }
+
+    override func setUpWithError() throws {
+        continueAfterFailure = false
+    }
+
+    @MainActor
+    func testLaunch() throws {
+        let app = XCUIApplication()
+        app.launch()
+
+        // Insert steps here to perform after app launch but before taking a screenshot,
+        // such as logging into a test account or navigating somewhere in the app
+
+        let attachment = XCTAttachment(screenshot: app.screenshot())
+        attachment.name = "Launch Screen"
+        attachment.lifetime = .keepAlways
+        add(attachment)
+    }
+}
@@ -0,0 +1,7 @@
+# Canary benchmark manifest
+#
+# Lists the suite files to include. Each file defines benchmarks
+# with shared constraints, topology, and default args.
+include = [
+    "single-m3-ultra.toml",
+]
@@ -0,0 +1,189 @@
+# Single-node M3 Ultra benchmarks
+#
+# Shared constraints applied to ALL benchmarks in this file.
+constraints = [
+    "All(MacOsBuild(=25D125))",
+    "Hosts(=1)",
+    "All(Chip(m3_ultra))",
+    "All(GpuCores(=80))",
+]
+
+[topology]
+type = "none"
+
+# Default args merged into each benchmark's args (benchmark-level args win).
+[defaults]
+pp = [512, 2048, 8192, 16384]
+tg = 128
+
+[[benchmark]]
+model = "mlx-community/Meta-Llama-3.1-70B-Instruct-4bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/gpt-oss-120b-MXFP4-Q8"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/GLM-4.7-Flash-8bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-Coder-Next-6bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-30B-A3B-8bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-0.6B-4bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-0.6B-8bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Llama-3.2-1B-Instruct-4bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Llama-3.2-3B-Instruct-4bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Llama-3.2-3B-Instruct-8bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Meta-Llama-3.1-8B-Instruct-4bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Meta-Llama-3.1-8B-Instruct-8bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Meta-Llama-3.1-8B-Instruct-bf16"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/gpt-oss-20b-MXFP4-Q8"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-30B-A3B-4bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/GLM-4.7-Flash-4bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/GLM-4.7-Flash-5bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/GLM-4.7-Flash-6bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Llama-3.3-70B-Instruct-4bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-Coder-Next-4bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-Coder-Next-5bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-Coder-Next-8bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-Next-80B-A3B-Instruct-4bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-Next-80B-A3B-Instruct-8bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-Next-80B-A3B-Thinking-4bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-Next-80B-A3B-Thinking-8bit"
+extra_constraints = ["All(Memory(>=96GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Llama-3.3-70B-Instruct-8bit"
+extra_constraints = ["All(Memory(>=256GiB))"]
+
+[[benchmark]]
+model = "mlx-community/llama-3.3-70b-instruct-fp16"
+extra_constraints = ["All(Memory(>=256GiB))"]
+
+[[benchmark]]
+model = "mlx-community/GLM-4.5-Air-8bit"
+extra_constraints = ["All(Memory(>=256GiB))"]
+
+[[benchmark]]
+model = "mlx-community/GLM-4.5-Air-bf16"
+extra_constraints = ["All(Memory(>=256GiB))"]
+
+[[benchmark]]
+model = "mlx-community/GLM-4.7-4bit"
+extra_constraints = ["All(Memory(>=256GiB))"]
+
+[[benchmark]]
+model = "mlx-community/MiniMax-M2.1-3bit"
+extra_constraints = ["All(Memory(>=256GiB))"]
+
+[[benchmark]]
+model = "mlx-community/MiniMax-M2.1-8bit"
+extra_constraints = ["All(Memory(>=256GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-235B-A22B-Instruct-2507-4bit"
+extra_constraints = ["All(Memory(>=256GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-Coder-Next-bf16"
+extra_constraints = ["All(Memory(>=256GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Step-3.5-Flash-4bit"
+extra_constraints = ["All(Memory(>=256GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Step-3.5-Flash-6bit"
+extra_constraints = ["All(Memory(>=256GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Step-3.5-Flash-8Bit"
+extra_constraints = ["All(Memory(>=256GiB))"]
+
+[[benchmark]]
+model = "mlx-community/DeepSeek-V3.1-4bit"
+extra_constraints = ["All(Memory(>=512GiB))"]
+
+[[benchmark]]
+model = "mlx-community/GLM-4.7-6bit"
+extra_constraints = ["All(Memory(>=512GiB))"]
+
+[[benchmark]]
+model = "mlx-community/GLM-4.7-8bit-gs32"
+extra_constraints = ["All(Memory(>=512GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-235B-A22B-Instruct-2507-8bit"
+extra_constraints = ["All(Memory(>=512GiB))"]
+
+[[benchmark]]
+model = "mlx-community/Qwen3-Coder-480B-A35B-Instruct-4bit"
+extra_constraints = ["All(Memory(>=512GiB))"]
@@ -186,7 +186,9 @@
  // topology payload is missing them. Topology order is preserved exactly so
  // that the mini preview matches the main TopologyGraph layout.
  const nodeList = $derived(() => {
-    const nodesFromTopology = Object.keys(nodes).map((id) => {
+    const nodesFromTopology = Object.keys(nodes)
+      .filter((id) => nodes[id].node_type !== "lite")
+      .map((id) => {
      const info = nodes[id];
      const totalBytes =
        info.macmon_info?.memory?.ram_total ?? info.system_info?.memory ?? 0;
@@ -554,6 +554,7 @@
      const clipPathId = `clip-${nodeInfo.id.replace(/[^a-zA-Z0-9]/g, "-")}`;

      const modelLower = modelId.toLowerCase();
+      const isLiteNode = node.node_type === "lite";

      // Check node states for styling
      const isHighlighted = highlightedNodes.has(nodeInfo.id);
@@ -906,6 +907,94 @@
          .attr("height", trackpadHeight)
          .attr("fill", "rgba(255,255,255,0.08)")
          .attr("rx", 2);
+      } else if (modelLower.includes("iphone")) {
+        // iPhone - rounded rectangle phone shape with Dynamic Island
+        iconBaseWidth = nodeRadius * 0.7;
+        iconBaseHeight = nodeRadius * 1.4;
+        const x = nodeInfo.x - iconBaseWidth / 2;
+        const y = nodeInfo.y - iconBaseHeight / 2;
+        const cornerRadius = iconBaseWidth * 0.2;
+        const bezel = 3;
+
+        // Phone body (outer frame)
+        nodeG
+          .append("rect")
+          .attr("class", "node-outline")
+          .attr("x", x)
+          .attr("y", y)
+          .attr("width", iconBaseWidth)
+          .attr("height", iconBaseHeight)
+          .attr("rx", cornerRadius)
+          .attr("fill", "#1a1a1a")
+          .attr("stroke", wireColor)
+          .attr("stroke-width", strokeWidth);
+
+        // Screen area (inner)
+        const screenClipId = `iphone-clip-${nodeInfo.id.replace(/[^a-zA-Z0-9]/g, "-")}`;
+        defs
+          .append("clipPath")
+          .attr("id", screenClipId)
+          .append("rect")
+          .attr("x", x + bezel)
+          .attr("y", y + bezel)
+          .attr("width", iconBaseWidth - bezel * 2)
+          .attr("height", iconBaseHeight - bezel * 2)
+          .attr("rx", cornerRadius - 1);
+
+        nodeG
+          .append("rect")
+          .attr("x", x + bezel)
+          .attr("y", y + bezel)
+          .attr("width", iconBaseWidth - bezel * 2)
+          .attr("height", iconBaseHeight - bezel * 2)
+          .attr("rx", cornerRadius - 1)
+          .attr("fill", screenFill);
+
+        // Memory fill on screen (fills from bottom up)
+        if (ramUsagePercent > 0) {
+          const memFillTotalHeight = iconBaseHeight - bezel * 2;
+          const memFillActualHeight =
+            (ramUsagePercent / 100) * memFillTotalHeight;
+          nodeG
+            .append("rect")
+            .attr("x", x + bezel)
+            .attr(
+              "y",
+              y + bezel + (memFillTotalHeight - memFillActualHeight),
+            )
+            .attr("width", iconBaseWidth - bezel * 2)
+            .attr("height", memFillActualHeight)
+            .attr("fill", "rgba(255,215,0,0.85)")
+            .attr("clip-path", `url(#${screenClipId})`);
+        }
+
+        // Dynamic Island notch (centered near top)
+        const diWidth = iconBaseWidth * 0.3;
+        const diHeight = iconBaseHeight * 0.04;
+        nodeG
+          .append("rect")
+          .attr("x", nodeInfo.x - diWidth / 2)
+          .attr("y", y + bezel + 4)
+          .attr("width", diWidth)
+          .attr("height", diHeight)
+          .attr("rx", diHeight / 2)
+          .attr("fill", "#000000");
+
+        // Apple logo on screen (centered)
+        const targetLogoHeight = iconBaseHeight * 0.14;
+        const logoScale = targetLogoHeight / LOGO_NATIVE_HEIGHT;
+        const logoX = nodeInfo.x - (LOGO_NATIVE_WIDTH * logoScale) / 2;
+        const logoY =
+          nodeInfo.y - (LOGO_NATIVE_HEIGHT * logoScale) / 2;
+        nodeG
+          .append("path")
+          .attr("d", APPLE_LOGO_PATH)
+          .attr(
+            "transform",
+            `translate(${logoX}, ${logoY}) scale(${logoScale})`,
+          )
+          .attr("fill", "#FFFFFF")
+          .attr("opacity", 0.9);
      } else {
        // Default/Unknown - holographic hexagon
        const hexRadius = nodeRadius * 0.6;
@@ -924,9 +1013,43 @@
          .attr("stroke-width", strokeWidth);
      }

+      // --- LITE badge for lite nodes ---
+      if (isLiteNode) {
+        const badgeX = nodeInfo.x + iconBaseWidth / 2 + 4;
+        const badgeY = nodeInfo.y - iconBaseHeight / 2 - 2;
+        const badgeFontSize = Math.max(9, nodeRadius * 0.12);
+        const badgePadH = 4;
+        const badgePadV = 2;
+        const badgeWidth = badgeFontSize * 2.8 + badgePadH * 2;
+        const badgeHeight = badgeFontSize + badgePadV * 2;
+
+        nodeG
+          .append("rect")
+          .attr("x", badgeX)
+          .attr("y", badgeY)
+          .attr("width", badgeWidth)
+          .attr("height", badgeHeight)
+          .attr("rx", 3)
+          .attr("fill", "rgba(255,215,0,0.15)")
+          .attr("stroke", "rgba(255,215,0,0.6)")
+          .attr("stroke-width", 1);
+
+        nodeG
+          .append("text")
+          .attr("x", badgeX + badgeWidth / 2)
+          .attr("y", badgeY + badgeHeight / 2)
+          .attr("text-anchor", "middle")
+          .attr("dominant-baseline", "central")
+          .attr("fill", "rgba(255,215,0,0.9)")
+          .attr("font-size", badgeFontSize)
+          .attr("font-weight", "700")
+          .attr("font-family", "SF Mono, Monaco, monospace")
+          .text("LITE");
+      }
+
      // --- Vertical GPU Bar (right side of icon) ---
-      // Show in both full mode and minimized mode (scaled appropriately)
-      if (showFullLabels || isMinimized) {
+      // Show in both full mode and minimized mode (scaled appropriately), but not for lite nodes
+      if ((showFullLabels || isMinimized) && !isLiteNode) {
        const gpuBarWidth = isMinimized
          ? Math.max(16, nodeRadius * 0.32)
          : Math.max(28, nodeRadius * 0.3);
@@ -50,6 +50,7 @@ export interface NodeInfo {
  last_macmon_update: number;
  friendly_name?: string;
  os_version?: string;
+  node_type?: string;
 }

 export interface TopologyEdge {
@@ -81,6 +82,7 @@ interface RawNodeIdentity {
  friendlyName?: string;
  osVersion?: string;
  osBuildVersion?: string;
+  nodeType?: string;
 }

 interface RawMemoryUsage {
@@ -445,6 +447,7 @@ function transformTopology(
      last_macmon_update: Date.now() / 1000,
      friendly_name: identity?.friendlyName,
      os_version: identity?.osVersion,
+      node_type: identity?.nodeType,
    };
  }

@@ -79,8 +79,9 @@
  const macosVersionMismatch = $derived.by(() => {
    if (!identitiesData) return null;
    const entries = Object.entries(identitiesData);
-    // Filter to macOS nodes (version starts with a digit, e.g. "15.3")
+    // Filter to full macOS nodes (version starts with a digit, e.g. "15.3"), excluding lite nodes
    const macosNodes = entries.filter(([_, id]) => {
+      if (id.nodeType === "lite") return false;
      const v = id.osVersion;
      return v && v !== "Unknown" && /^\d/.test(v);
    });
@@ -638,7 +639,9 @@
        models = data.data || [];
        // Restore last launch defaults if available
        const currentNodeCount = topologyData()
-          ? Object.keys(topologyData()!.nodes).length
+          ? Object.values(topologyData()!.nodes).filter(
+              (n) => n.node_type !== "lite",
+            ).length
          : 1;
        applyLaunchDefaults(models, currentNodeCount);
      }
@@ -1646,7 +1649,12 @@
    saveLaunchDefaults();
  }

-  const nodeCount = $derived(data ? Object.keys(data.nodes).length : 0);
+  const totalNodeCount = $derived(data ? Object.keys(data.nodes).length : 0);
+  const fullNodeCount = $derived(
+    data
+      ? Object.values(data.nodes).filter((n) => n.node_type !== "lite").length
+      : 0,
+  );
  const instanceCount = $derived(Object.keys(instanceData).length);

  // Helper to get the number of nodes in a placement preview
@@ -1659,7 +1667,7 @@
  }

  // Available min nodes options based on topology (like old dashboard)
-  const availableMinNodes = $derived(Math.max(1, nodeCount));
+  const availableMinNodes = $derived(Math.max(1, fullNodeCount));

  // Compute which min node values have valid previews for the current model/sharding/instance type
  // A minNodes value N is valid if there exists a placement with nodeCount >= N
@@ -1752,15 +1760,17 @@
  // Calculate total memory usage across all nodes
  const clusterMemory = $derived(() => {
    if (!data) return { used: 0, total: 0 };
-    return Object.values(data.nodes).reduce(
-      (acc, n) => {
-        const total =
-          n.macmon_info?.memory?.ram_total ?? n.system_info?.memory ?? 0;
-        const used = n.macmon_info?.memory?.ram_usage ?? 0;
-        return { used: acc.used + used, total: acc.total + total };
-      },
-      { used: 0, total: 0 },
-    );
+    return Object.values(data.nodes)
+      .filter((n) => n.node_type !== "lite")
+      .reduce(
+        (acc, n) => {
+          const total =
+            n.macmon_info?.memory?.ram_total ?? n.system_info?.memory ?? 0;
+          const used = n.macmon_info?.memory?.ram_usage ?? 0;
+          return { used: acc.used + used, total: acc.total + total };
+        },
+        { used: 0, total: 0 },
+      );
  });
 </script>

@@ -3091,7 +3101,7 @@
                  TOPOLOGY
                </div>
                <span class="text-xs text-white/70 tabular-nums"
-                  >{nodeCount} {nodeCount === 1 ? "NODE" : "NODES"}</span
+                  >{totalNodeCount} {totalNodeCount === 1 ? "NODE" : "NODES"}</span
                >
              </div>

@@ -132,7 +132,7 @@ markers = [
 env = [
  "EXO_TESTS=1"
 ]
-addopts = "-m 'not slow'"
+addopts = "-m 'not slow' --ignore=tests/start_distributed_test.py"
 filterwarnings = [
    "ignore:builtin type Swig:DeprecationWarning",
 ]
@@ -14,6 +14,7 @@ from exo.download.download_utils import (
    map_repo_download_progress_to_download_progress_data,
 )
 from exo.download.shard_downloader import ShardDownloader
+from exo.shared.constants import EXO_MODELS_DIR
 from exo.shared.models.model_cards import ModelId
 from exo.shared.types.commands import (
    CancelDownload,
@@ -63,6 +64,9 @@ class DownloadCoordinator:
        self.event_sender, self.event_receiver = channel[Event]()
        self.shard_downloader.on_progress(self._download_progress_callback)

+    def _model_dir(self, model_id: ModelId) -> str:
+        return str(EXO_MODELS_DIR / model_id.normalize())
+
    async def _download_progress_callback(
        self, callback_shard: ShardMetadata, progress: RepoDownloadProgress
    ) -> None:
@@ -74,6 +78,7 @@ class DownloadCoordinator:
                shard_metadata=callback_shard,
                node_id=self.node_id,
                total_bytes=progress.total_bytes,
+                model_directory=self._model_dir(model_id),
            )
            self.download_status[model_id] = completed
            await self.event_sender.send(
@@ -93,6 +98,7 @@ class DownloadCoordinator:
                download_progress=map_repo_download_progress_to_download_progress_data(
                    progress
                ),
+                model_directory=self._model_dir(model_id),
            )
            self.download_status[model_id] = ongoing
            await self.event_sender.send(
@@ -170,7 +176,11 @@ class DownloadCoordinator:
                return

        # Emit pending status
-        progress = DownloadPending(shard_metadata=shard, node_id=self.node_id)
+        progress = DownloadPending(
+            shard_metadata=shard,
+            node_id=self.node_id,
+            model_directory=self._model_dir(model_id),
+        )
        self.download_status[model_id] = progress
        await self.event_sender.send(NodeDownloadProgress(download_progress=progress))

@@ -184,6 +194,7 @@ class DownloadCoordinator:
                shard_metadata=shard,
                node_id=self.node_id,
                total_bytes=initial_progress.total_bytes,
+                model_directory=self._model_dir(model_id),
            )
            self.download_status[model_id] = completed
            await self.event_sender.send(
@@ -206,6 +217,7 @@ class DownloadCoordinator:
            download_progress=map_repo_download_progress_to_download_progress_data(
                initial_progress
            ),
+            model_directory=self._model_dir(model_id),
        )
        self.download_status[model_id] = status
        self.event_sender.send_nowait(NodeDownloadProgress(download_progress=status))
@@ -219,6 +231,7 @@ class DownloadCoordinator:
                    shard_metadata=shard,
                    node_id=self.node_id,
                    error_message=str(e),
+                    model_directory=self._model_dir(model_id),
                )
                self.download_status[model_id] = failed
                await self.event_sender.send(
@@ -253,6 +266,7 @@ class DownloadCoordinator:
            pending = DownloadPending(
                shard_metadata=current_status.shard_metadata,
                node_id=self.node_id,
+                model_directory=self._model_dir(model_id),
            )
            await self.event_sender.send(
                NodeDownloadProgress(download_progress=pending)
@@ -295,11 +309,18 @@ class DownloadCoordinator:
                            node_id=self.node_id,
                            shard_metadata=progress.shard,
                            total_bytes=progress.total_bytes,
+                            model_directory=self._model_dir(
+                                progress.shard.model_card.model_id
+                            ),
                        )
                    elif progress.status in ["in_progress", "not_started"]:
                        if progress.downloaded_bytes_this_session.in_bytes == 0:
                            status = DownloadPending(
-                                node_id=self.node_id, shard_metadata=progress.shard
+                                node_id=self.node_id,
+                                shard_metadata=progress.shard,
+                                model_directory=self._model_dir(
+                                    progress.shard.model_card.model_id
+                                ),
                            )
                        else:
                            status = DownloadOngoing(
@@ -308,6 +329,9 @@ class DownloadCoordinator:
                                download_progress=map_repo_download_progress_to_download_progress_data(
                                    progress
                                ),
+                                model_directory=self._model_dir(
+                                    progress.shard.model_card.model_id
+                                ),
                            )
                    else:
                        continue
@@ -144,8 +144,8 @@ async def collect_responses_response(
            for tool in chunk.tool_calls:
                function_call_items.append(
                    ResponseFunctionCallItem(
-                        id=f"fc_{tool.id}",
-                        call_id=f"call_{tool.id}",
+                        id=tool.id,
+                        call_id=tool.id,
                        name=tool.name,
                        arguments=tool.arguments,
                    )
@@ -85,6 +85,7 @@ from exo.shared.types.api import (
    ImageGenerationTaskParams,
    ImageListItem,
    ImageListResponse,
+    LiteNodeHeartbeatRequest,
    ModelList,
    ModelListModel,
    PlaceInstanceParams,
@@ -122,6 +123,7 @@ from exo.shared.types.commands import (
    ForwarderDownloadCommand,
    ImageEdits,
    ImageGeneration,
+    LiteNodeHeartbeat,
    PlaceInstance,
    SendInputChunk,
    StartDownload,
@@ -148,6 +150,7 @@ from exo.shared.types.worker.shards import Sharding
 from exo.utils.banner import print_startup_banner
 from exo.utils.channels import Receiver, Sender, channel
 from exo.utils.event_buffer import OrderedBuffer
+from exo.utils.info_gatherer.info_gatherer import LiteNodeRegistration

 _API_EVENT_LOG_DIR = EXO_EVENT_LOG_DIR / "api"

@@ -303,6 +306,7 @@ class API:
        self.app.get("/v1/traces/{task_id}")(self.get_trace)
        self.app.get("/v1/traces/{task_id}/stats")(self.get_trace_stats)
        self.app.get("/v1/traces/{task_id}/raw")(self.get_trace_raw)
+        self.app.post("/v1/lite_node/heartbeat")(self.lite_node_heartbeat)

    async def place_instance(self, payload: PlaceInstanceParams):
        command = PlaceInstance(
@@ -365,6 +369,7 @@ class API:
                node_network=self.state.node_network,
                topology=self.state.topology,
                current_instances=self.state.instances,
+                node_identities=self.state.node_identities,
            )
        except ValueError as exc:
            raise HTTPException(status_code=400, detail=str(exc)) from exc
@@ -390,7 +395,14 @@ class API:
        previews: list[PlacementPreview] = []
        required_nodes = set(node_ids) if node_ids else None

-        if len(list(self.state.topology.list_nodes())) == 0:
+        full_node_count = sum(
+            1
+            for nid in self.state.topology.list_nodes()
+            if self.state.node_identities.get(nid) is None
+            or self.state.node_identities[nid].node_type != "lite"
+        )
+
+        if full_node_count == 0:
            return PlacementPreviewResponse(previews=[])

        try:
@@ -405,9 +417,7 @@ class API:
                instance_combinations.extend(
                    [
                        (sharding, instance_meta, i)
-                        for i in range(
-                            1, len(list(self.state.topology.list_nodes())) + 1
-                        )
+                        for i in range(1, full_node_count + 1)
                    ]
                )
        # TODO: PDD
@@ -427,6 +437,7 @@ class API:
                    topology=self.state.topology,
                    current_instances=self.state.instances,
                    required_nodes=required_nodes,
+                    node_identities=self.state.node_identities,
                )
            except ValueError as exc:
                if (model_card.model_id, sharding, instance_meta, 0) not in seen:
@@ -1278,11 +1289,30 @@ class API:
                media_type="application/json",
            )

+    async def lite_node_heartbeat(self, payload: LiteNodeHeartbeatRequest) -> JSONResponse:
+        info = LiteNodeRegistration(
+            model=payload.model,
+            chip=payload.chip,
+            os_version=payload.os_version,
+            friendly_name=payload.friendly_name,
+            ram_total=payload.ram_total,
+            ram_available=payload.ram_available,
+        )
+        command = LiteNodeHeartbeat(
+            target_node_id=NodeId(payload.node_id),
+            info=info,
+        )
+        await self._send(command)
+        return JSONResponse({"status": "ok"})
+
    def _calculate_total_available_memory(self) -> Memory:
-        """Calculate total available memory across all nodes in bytes."""
+        """Calculate total available memory across all non-lite nodes in bytes."""
        total_available = Memory()

-        for memory in self.state.node_memory.values():
+        for node_id, memory in self.state.node_memory.items():
+            identity = self.state.node_identities.get(node_id)
+            if identity is not None and identity.node_type == "lite":
+                continue
            total_available += memory.ram_available

        return total_available
@@ -1367,6 +1397,7 @@ class API:
    async def run(self):
        shutdown_ev = anyio.Event()

+        bonjour_cleanup = self._register_bonjour_service()
        try:
            async with create_task_group() as tg:
                self._tg = tg
@@ -1382,10 +1413,48 @@ class API:
                    with anyio.CancelScope(shield=True):
                        shutdown_ev.set()
        finally:
+            bonjour_cleanup()
            self._event_log.close()
            self.command_sender.close()
            self.global_event_receiver.close()

+    def _register_bonjour_service(self) -> Callable[[], None]:
+        """Register a Bonjour service via the system mDNSResponder. Returns a cleanup function."""
+        import subprocess
+        import sys
+
+        if sys.platform != "darwin":
+            logger.info("Bonjour service registration is only supported on macOS")
+            return lambda: None
+
+        service_name = f"EXO Cluster ({self.node_id[:8]})"
+        try:
+            proc = subprocess.Popen(
+                [
+                    "dns-sd",
+                    "-R",
+                    service_name,
+                    "_exo._tcp",
+                    "local",
+                    str(self.port),
+                    f"node_id={self.node_id}",
+                ],
+                stdout=subprocess.DEVNULL,
+                stderr=subprocess.DEVNULL,
+            )
+            logger.info(
+                f"Registered Bonjour service _exo._tcp on port {self.port} (pid {proc.pid})"
+            )
+
+            def cleanup() -> None:
+                proc.terminate()
+                proc.wait()
+
+            return cleanup
+        except Exception as e:
+            logger.warning(f"Failed to register Bonjour service: {e}")
+            return lambda: None
+
    async def run_api(self, ev: anyio.Event):
        cfg = Config()
        cfg.bind = [f"0.0.0.0:{self.port}"]
@@ -21,6 +21,7 @@ from exo.shared.types.commands import (
    ForwarderDownloadCommand,
    ImageEdits,
    ImageGeneration,
+    LiteNodeHeartbeat,
    PlaceInstance,
    RequestEventLog,
    SendInputChunk,
@@ -299,6 +300,7 @@ class Master:
                                self.state.instances,
                                self.state.node_memory,
                                self.state.node_network,
+                                node_identities=self.state.node_identities,
                            )
                            transition_events = get_transition_events(
                                self.state.instances, placement, self.state.tasks
@@ -344,6 +346,14 @@ class Master:
                            self.command_task_mapping.pop(
                                command.finished_command_id, None
                            )
+                        case LiteNodeHeartbeat():
+                            generated_events.append(
+                                NodeGatheredInfo(
+                                    node_id=command.target_node_id,
+                                    when=str(datetime.now(tz=timezone.utc)),
+                                    info=command.info,
+                                )
+                            )
                        case RequestEventLog():
                            # We should just be able to send everything, since other buffers will ignore old messages
                            # rate limit to 1000 at a time
@@ -29,7 +29,7 @@ from exo.shared.types.events import (
    TaskStatusUpdated,
 )
 from exo.shared.types.memory import Memory
-from exo.shared.types.profiling import MemoryUsage, NodeNetworkInfo
+from exo.shared.types.profiling import MemoryUsage, NodeIdentity, NodeNetworkInfo
 from exo.shared.types.tasks import Task, TaskId, TaskStatus
 from exo.shared.types.worker.downloads import (
    DownloadOngoing,
@@ -67,8 +67,21 @@ def place_instance(
    node_memory: Mapping[NodeId, MemoryUsage],
    node_network: Mapping[NodeId, NodeNetworkInfo],
    required_nodes: set[NodeId] | None = None,
+    node_identities: Mapping[NodeId, NodeIdentity] = {},
 ) -> dict[InstanceId, Instance]:
+    lite_node_ids = {
+        nid
+        for nid, identity in node_identities.items()
+        if identity.node_type == "lite"
+    }
    cycles = topology.get_cycles()
+    if lite_node_ids:
+        filtered_cycles: list[Cycle] = []
+        for cycle in cycles:
+            filtered_nodes = [nid for nid in cycle.node_ids if nid not in lite_node_ids]
+            if filtered_nodes:
+                filtered_cycles.append(Cycle(filtered_nodes))
+        cycles = filtered_cycles
    candidate_cycles = list(filter(lambda it: len(it) >= command.min_nodes, cycles))

    # Filter to cycles containing all required nodes (subset matching)
@@ -42,6 +42,7 @@ from exo.shared.types.worker.downloads import DownloadProgress
 from exo.shared.types.worker.instances import Instance, InstanceId
 from exo.shared.types.worker.runners import RunnerId, RunnerStatus
 from exo.utils.info_gatherer.info_gatherer import (
+    LiteNodeRegistration,
    MacmonMetrics,
    MacThunderboltConnections,
    MacThunderboltIdentifiers,
@@ -218,11 +219,6 @@ def apply_node_timed_out(event: NodeTimedOut, state: State) -> State:
        key: value for key, value in state.downloads.items() if key != event.node_id
    }
    # Clean up all granular node mappings
-    node_identities = {
-        key: value
-        for key, value in state.node_identities.items()
-        if key != event.node_id
-    }
    node_memory = {
        key: value for key, value in state.node_memory.items() if key != event.node_id
    }
@@ -248,6 +244,11 @@ def apply_node_timed_out(event: NodeTimedOut, state: State) -> State:
    node_rdma_ctl = {
        key: value for key, value in state.node_rdma_ctl.items() if key != event.node_id
    }
+    node_identities = {
+        key: value
+        for key, value in state.node_identities.items()
+        if key != event.node_id
+    }
    # Only recompute cycles if the leaving node had TB bridge enabled
    leaving_node_status = state.node_thunderbolt_bridge.get(event.node_id)
    leaving_node_had_tb_enabled = (
@@ -263,7 +264,6 @@ def apply_node_timed_out(event: NodeTimedOut, state: State) -> State:
            "downloads": downloads,
            "topology": topology,
            "last_seen": last_seen,
-            "node_identities": node_identities,
            "node_memory": node_memory,
            "node_disk": node_disk,
            "node_system": node_system,
@@ -271,6 +271,7 @@ def apply_node_timed_out(event: NodeTimedOut, state: State) -> State:
            "node_thunderbolt": node_thunderbolt,
            "node_thunderbolt_bridge": node_thunderbolt_bridge,
            "node_rdma_ctl": node_rdma_ctl,
+            "node_identities": node_identities,
            "thunderbolt_bridge_cycles": thunderbolt_bridge_cycles,
        }
    )
@@ -377,6 +378,30 @@ def apply_node_gathered_info(event: NodeGatheredInfo, state: State) -> State:
                **state.node_rdma_ctl,
                event.node_id: NodeRdmaCtlStatus(enabled=info.enabled),
            }
+        case LiteNodeRegistration():
+            current_identity = state.node_identities.get(event.node_id, NodeIdentity())
+            new_identity = current_identity.model_copy(
+                update={
+                    "model_id": info.model,
+                    "chip_id": info.chip,
+                    "os_version": info.os_version,
+                    "friendly_name": info.friendly_name,
+                    "node_type": "lite",
+                }
+            )
+            update["node_identities"] = {
+                **state.node_identities,
+                event.node_id: new_identity,
+            }
+            update["node_memory"] = {
+                **state.node_memory,
+                event.node_id: MemoryUsage.from_bytes(
+                    ram_total=info.ram_total,
+                    ram_available=info.ram_available,
+                    swap_total=0,
+                    swap_available=0,
+                ),
+            }

    return state.model_copy(update=update)

@@ -406,3 +406,13 @@ class TraceListItem(CamelCaseModel):

 class TraceListResponse(CamelCaseModel):
    traces: list[TraceListItem]
+
+
+class LiteNodeHeartbeatRequest(CamelCaseModel):
+    node_id: str
+    model: str
+    chip: str
+    os_version: str
+    friendly_name: str
+    ram_total: int
+    ram_available: int
@@ -10,6 +10,7 @@ from exo.shared.types.common import CommandId, NodeId
 from exo.shared.types.text_generation import TextGenerationTaskParams
 from exo.shared.types.worker.instances import Instance, InstanceId, InstanceMeta
 from exo.shared.types.worker.shards import Sharding, ShardMetadata
+from exo.utils.info_gatherer.info_gatherer import GatheredInfo
 from exo.utils.pydantic_ext import CamelCaseModel, TaggedModel


@@ -62,6 +63,11 @@ class SendInputChunk(BaseCommand):
    chunk: InputImageChunk


+class LiteNodeHeartbeat(BaseCommand):
+    target_node_id: NodeId
+    info: GatheredInfo
+
+
 class RequestEventLog(BaseCommand):
    since_idx: int

@@ -96,6 +102,7 @@ Command = (
    | TaskCancelled
    | TaskFinished
    | SendInputChunk
+    | LiteNodeHeartbeat
 )


@@ -68,6 +68,8 @@ class SystemPerformanceProfile(CamelCaseModel):

 InterfaceType = Literal["wifi", "ethernet", "maybe_ethernet", "thunderbolt", "unknown"]

+NodeType = Literal["full", "lite"]
+

 class NetworkInterfaceInfo(CamelCaseModel):
    name: str
@@ -83,6 +85,7 @@ class NodeIdentity(CamelCaseModel):
    friendly_name: str = "Unknown"
    os_version: str = "Unknown"
    os_build_version: str = "Unknown"
+    node_type: NodeType = "full"


 class NodeNetworkInfo(CamelCaseModel):
@@ -26,6 +26,7 @@ class DownloadProgressData(CamelCaseModel):
 class BaseDownloadProgress(TaggedModel):
    node_id: NodeId
    shard_metadata: ShardMetadata
+    model_directory: str = ""


 class DownloadPending(BaseDownloadProgress):
@@ -1,5 +1,7 @@
+import sys
+
+
 def print_startup_banner(port: int) -> None:
-    """Print a prominent startup banner with API endpoint information."""
    dashboard_url = f"http://localhost:{port}"
    banner = f"""
 ╔═══════════════════════════════════════════════════════════════════════╗
@@ -27,4 +29,4 @@ def print_startup_banner(port: int) -> None:

 """

-    print(banner)
+    print(banner, file=sys.stderr)
@@ -334,6 +334,17 @@ class NodeDiskUsage(TaggedModel):
        )


+class LiteNodeRegistration(TaggedModel):
+    """Device info reported by a lite node (e.g. iPhone) via heartbeat."""
+
+    model: str  # e.g. "iPhone 16 Pro"
+    chip: str  # e.g. "Apple A18 Pro"
+    os_version: str  # e.g. "18.3"
+    friendly_name: str  # e.g. "Sami's iPhone"
+    ram_total: int  # bytes
+    ram_available: int  # bytes
+
+
 async def _gather_iface_map() -> dict[str, str] | None:
    proc = await anyio.run_process(
        ["networksetup", "-listallhardwareports"], check=False
@@ -366,6 +377,7 @@ GatheredInfo = (
    | MiscData
    | StaticNodeInformation
    | NodeDiskUsage
+    | LiteNodeRegistration
 )


@@ -306,7 +306,7 @@ def mlx_generate(
    max_stop_len = max((len(s) for s in stop_sequences), default=0)

    mx_barrier(group)
-    logger.info("Ready to prefill")
+    logger.info("Starting prefill")

    # Prefill cache with all tokens except the last one
    prefill_tps, prefill_tokens, ssm_snapshots_list = prefill(
@@ -353,7 +353,13 @@ def load_tokenizer_for_model_id(
            return list(hf_tokenizer.model.encode(text, allowed_special="all"))  # pyright: ignore[reportUnknownMemberType,reportUnknownArgumentType]

        hf_tokenizer.encode = _patched_encode
-        return TokenizerWrapper(hf_tokenizer, eos_token_ids=eos_token_ids)
+        return TokenizerWrapper(
+            hf_tokenizer,
+            eos_token_ids=eos_token_ids,
+            tool_call_start="<|tool_calls_section_begin|>",
+            tool_call_end="<|tool_calls_section_end|>",
+            tool_parser=_parse_kimi_tool_calls,
+        )

    tokenizer = load_tokenizer(
        model_path,
@@ -585,3 +591,41 @@ def mx_barrier(group: Group | None):
            mx.array(1.0), group=group, stream=mx.default_stream(mx.Device(mx.cpu))
        )
    )
+
+
+def _parse_kimi_tool_calls(text: str):
+    import regex as re
+
+    # kimi has a fixed function naming scheme, with a json formatted arg
+    #   functions.multiply:0<|tool_call_argument_begin|>{"a": 2, "b": 3}
+    _func_name_regex = re.compile(
+        r"^\s*((?:functions\.)?(.+?):\d+)\s*<\|tool_call_argument_begin\|>", re.DOTALL
+    )
+    _func_arg_regex = re.compile(r"<\|tool_call_argument_begin\|>\s*(.*)\s*", re.DOTALL)
+    _tool_call_split_regex = re.compile(
+        r"<\|tool_call_begin\|>(.*?)<\|tool_call_end\|>", re.DOTALL
+    )
+
+    def _parse_single_tool(text: str) -> dict[str, Any]:
+        func_name_match = _func_name_regex.search(text)
+        if func_name_match is None:
+            raise ValueError("No tool call found.")
+        tool_call_id = func_name_match.group(1)  # e.g. "functions.get_weather:0"
+        func_name = func_name_match.group(2)  # e.g. "get_weather"
+
+        func_args_match = _func_arg_regex.search(text)
+        if func_args_match is None:
+            raise ValueError("No tool call arguments found.")
+        func_args = func_args_match.group(1)
+        try:
+            arg_dct = json.loads(func_args)  # pyright: ignore[reportAny]
+        except Exception:
+            arg_dct = None
+
+        return dict(id=tool_call_id, name=func_name, arguments=arg_dct)
+
+    tool_matches = _tool_call_split_regex.findall(text)
+    if tool_matches:
+        return [_parse_single_tool(match) for match in tool_matches]  # pyright: ignore[reportAny]
+    else:
+        return [_parse_single_tool(text)]
@@ -1,11 +1,10 @@
 import base64
-import json
 import math
 import resource
 import time
 from collections.abc import Generator
 from functools import cache
-from typing import Any, Callable, Literal
+from typing import Literal

 import mlx.core as mx
 from mlx_lm.models.gpt_oss import Model as GptOssModel
@@ -16,7 +15,6 @@ from openai_harmony import (  # pyright: ignore[reportMissingTypeStubs]
    StreamableParser,
    load_harmony_encoding,
 )
-from pydantic import ValidationError

 from exo.shared.constants import EXO_MAX_CHUNK_SIZE, EXO_TRACING_ENABLED
 from exo.shared.models.model_cards import ModelId, ModelTask
@@ -93,6 +91,8 @@ from exo.worker.engines.mlx.utils_mlx import (
 )
 from exo.worker.runner.bootstrap import logger

+from .tool_parsers import ToolParser, make_mlx_parser
+

 def _is_primary_output_node(shard_metadata: ShardMetadata) -> bool:
    """Check if this node is the primary output node for image generation.
@@ -138,6 +138,7 @@ def main(
    inference_model: Model | None = None
    image_model: DistributedImageModel | None = None
    tokenizer = None
+    tool_parser: ToolParser | None = None
    group = None
    kv_prefix_cache: KVPrefixCache | None = None
    check_for_cancel_every: int | None = None
@@ -203,8 +204,17 @@ def main(
                            bound_instance, group, on_timeout=on_model_load_timeout
                        )
                        logger.info(
-                            f"model has_tool_calling={tokenizer.has_tool_calling}"
+                            f"model has_tool_calling={tokenizer.has_tool_calling} using tokens {tokenizer.tool_call_start}, {tokenizer.tool_call_end}"
                        )
+                        if tokenizer.has_tool_calling:
+                            assert tokenizer.tool_call_start
+                            assert tokenizer.tool_call_end
+                            assert tokenizer.tool_parser  # pyright: ignore[reportAny]
+                            tool_parser = make_mlx_parser(
+                                tokenizer.tool_call_start,
+                                tokenizer.tool_call_end,
+                                tokenizer.tool_parser,  # pyright: ignore[reportAny]
+                            )
                        kv_prefix_cache = KVPrefixCache(group)

                    elif (
@@ -310,31 +320,11 @@ def main(
                                mlx_generator, tokenizer
                            )

-                        # Kimi-K2 has tool call sections - we don't care about them
-                        if "kimi" in shard_metadata.model_card.model_id.lower():
-                            mlx_generator = filter_kimi_tokens(mlx_generator)
-                            patch_kimi_tokenizer(tokenizer)
-
-                        # GLM models need patched parser (upstream has bug with None regex match)
-                        elif "glm" in shard_metadata.model_card.model_id.lower():
-                            patch_glm_tokenizer(tokenizer)
-
                        # GPT-OSS specific parsing to match other model formats.
-                        elif isinstance(inference_model, GptOssModel):
+                        if isinstance(inference_model, GptOssModel):
                            mlx_generator = parse_gpt_oss(mlx_generator)
-
-                        if tokenizer.has_tool_calling and not isinstance(
-                            inference_model, GptOssModel
-                        ):
-                            assert tokenizer.tool_call_start
-                            assert tokenizer.tool_call_end
-                            assert tokenizer.tool_parser  # pyright: ignore[reportAny]
-                            mlx_generator = parse_tool_calls(
-                                mlx_generator,
-                                tokenizer.tool_call_start,
-                                tokenizer.tool_call_end,
-                                tokenizer.tool_parser,  # pyright: ignore[reportAny]
-                            )
+                        elif tool_parser:
+                            mlx_generator = parse_tool_calls(mlx_generator, tool_parser)

                        completion_tokens = 0
                        tokens_since_last_cancel_check = 0
@@ -587,21 +577,8 @@ def get_gpt_oss_encoding():
    return encoding


-def filter_kimi_tokens(
-    responses: Generator[GenerationResponse | ToolCallResponse],
-) -> Generator[GenerationResponse]:
-    for resp in responses:
-        assert isinstance(resp, GenerationResponse)
-        if (
-            resp.text == "<|tool_calls_section_begin|>"
-            or resp.text == "<|tool_calls_section_end|>"
-        ):
-            continue
-        yield resp
-
-
 def parse_gpt_oss(
-    responses: Generator[GenerationResponse | ToolCallResponse],
+    responses: Generator[GenerationResponse],
 ) -> Generator[GenerationResponse | ToolCallResponse]:
    encoding = get_gpt_oss_encoding()
    stream = StreamableParser(encoding, role=Role.ASSISTANT)
@@ -658,9 +635,9 @@ def parse_gpt_oss(


 def parse_thinking_models(
-    responses: Generator[GenerationResponse | ToolCallResponse],
+    responses: Generator[GenerationResponse],
    tokenizer: TokenizerWrapper,
-) -> Generator[GenerationResponse | ToolCallResponse]:
+) -> Generator[GenerationResponse]:
    """
    For models that inject thinking tags in the prompt (like GLM-4.7),
    prepend the thinking tag to the output stream so the frontend
@@ -781,221 +758,55 @@ def _process_image_response(


 def parse_tool_calls(
-    responses: Generator[GenerationResponse | ToolCallResponse],
-    tool_call_start: str,
-    tool_call_end: str,
-    tool_parser: Callable[[str], dict[str, Any] | list[dict[str, Any]]],
+    responses: Generator[GenerationResponse], tool_parser: ToolParser
 ) -> Generator[GenerationResponse | ToolCallResponse]:
    in_tool_call = False
    tool_call_text_parts: list[str] = []
    for response in responses:
-        assert isinstance(response, GenerationResponse)
-        # assumption: the tool call start is one token
-        if response.text == tool_call_start:
+        if response.text.startswith(tool_parser.start_parsing):
            in_tool_call = True
-            continue
-        # assumption: the tool call end is one token
-        if in_tool_call and response.text == tool_call_end:
-            try:
-                # tool_parser returns an arbitrarily nested python dictionary
-                # we actually don't want the python dictionary, we just want to
-                # parse the top level { function: ..., arguments: ... } structure
-                # as we're just gonna hand it back to the api anyway
-                parsed = tool_parser("".join(tool_call_text_parts).strip())
-                logger.info(f"parsed {tool_call_text_parts=} into {parsed=}")
-                if isinstance(parsed, list):
-                    tools = [_validate_single_tool(tool) for tool in parsed]
-                else:
-                    tools = [_validate_single_tool(parsed)]
-                yield ToolCallResponse(
-                    tool_calls=tools, usage=response.usage, stats=response.stats
-                )
-
-            except (
-                json.JSONDecodeError,
-                ValidationError,
-                ValueError,
-                AttributeError,
-            ) as e:
-                # ValueError: our parsers raise this for malformed tool calls
-                # AttributeError: upstream parsers (e.g. glm47) may raise this when regex doesn't match
-                logger.opt(exception=e).warning("tool call parsing failed")
-                # assumption: talking about tool calls, not making a tool call
-                response.text = (
-                    tool_call_start + "".join(tool_call_text_parts) + tool_call_end
-                )
-                yield response
-
-            in_tool_call = False
-            tool_call_text_parts = []
-            continue

        if in_tool_call:
            tool_call_text_parts.append(response.text)
+            if response.text.endswith(tool_parser.end_parsing):
+                # parse the actual tool calls from the tool call text
+                parsed = tool_parser.parse_tool_calls(
+                    "".join(tool_call_text_parts).strip()
+                )
+                logger.info(f"parsed {tool_call_text_parts=} into {parsed=}")
+                if parsed is not None:
+                    yield ToolCallResponse(
+                        tool_calls=parsed, usage=response.usage, stats=response.stats
+                    )
+                else:
+                    logger.warning(
+                        f"tool call parsing failed for text {''.join(tool_call_text_parts)}"
+                    )
+                    response.text = "".join(tool_call_text_parts)
+                    yield response
+
+                in_tool_call = False
+                tool_call_text_parts = []
+                continue
+
            if response.finish_reason is not None:
                logger.info(
-                    "toll call parsing interrupted, yield partial tool call as text"
+                    "tool call parsing interrupted, yield partial tool call as text"
                )
-                yield GenerationResponse(
-                    text=tool_call_start + "".join(tool_call_text_parts),
-                    token=0,
-                    finish_reason=response.finish_reason,
-                    usage=response.usage,
-                    stats=response.stats,
+                response = response.model_copy(
+                    update={
+                        "text": "".join(tool_call_text_parts),
+                        "token": 0,
+                    }
                )
+                yield response
+
            continue
+
        # fallthrough
        yield response


-def patch_kimi_tokenizer(tokenizer: TokenizerWrapper):
-    """
-    Version of to-be-upstreamed kimi-k2 tool parser
-    """
-    import ast
-    import json
-    from typing import Any
-
-    import regex as re
-
-    # kimi has a fixed function naming scheme, with a json formatted arg
-    #   functions.multiply:0 <|tool_call_argument_begin|> {"a": 2, "b": 3}
-    #   Also needs to handle tools like call_0<|tool_call_argument_begin|>{"filePath": "..."}
-    _func_name_regex = re.compile(
-        r"^\s*(.+)[:](\d+)\s*<\|tool_call_argument_begin\|>", re.DOTALL
-    )
-    _func_arg_regex = re.compile(r"<\|tool_call_argument_begin\|>\s*(.*)\s*", re.DOTALL)
-
-    # kimi has a tool_calls_section - we're leaving this up to the caller to handle
-    tool_call_start = "<|tool_call_begin|>"
-    tool_call_end = "<|tool_call_end|>"
-
-    def _deserialize(value: str) -> Any:  # pyright: ignore[reportAny]
-        try:
-            return json.loads(value)  # pyright: ignore[reportAny]
-        except Exception:
-            pass
-
-        try:
-            return ast.literal_eval(value)  # pyright: ignore[reportAny]
-        except Exception:
-            pass
-        return value
-
-    def parse_tool_call(text: str, tools: Any | None = None):
-        func_name_match = _func_name_regex.search(text)
-        if func_name_match is None:
-            raise ValueError(f"Could not parse function name from tool call: {text!r}")
-        original_func_name = func_name_match.group(1)
-        tool_id = func_name_match.group(2)
-        # strip off the `functions.` prefix, if it exists.
-        func_name = original_func_name[original_func_name.find(".") + 1 :]
-
-        func_args_match = _func_arg_regex.search(text)
-        if func_args_match is None:
-            raise ValueError(f"Could not parse function args from tool call: {text!r}")
-        func_args = func_args_match.group(1)
-        # the args should be valid json - no need to check against our tools to deserialize
-        arg_dct = _deserialize(func_args)  # pyright: ignore[reportAny]
-
-        return dict(
-            id=f"{original_func_name}:{tool_id}",
-            name=func_name,
-            arguments=arg_dct,  # pyright: ignore[reportAny]
-        )
-
-    tokenizer._tool_call_start = tool_call_start
-    tokenizer._tool_call_end = tool_call_end
-    tokenizer._tool_parser = parse_tool_call
-
-
-def patch_glm_tokenizer(tokenizer: TokenizerWrapper):
-    """
-    Fixed version of mlx_lm's glm47 tool parser that handles regex match failures.
-    """
-    import ast
-    import json
-    from typing import Any
-
-    import regex as re
-
-    _func_name_regex = re.compile(r"^(.*?)<arg_key>", re.DOTALL)
-    _func_arg_regex = re.compile(
-        r"<arg_key>(.*?)</arg_key>(?:\n|\s)*<arg_value>(.*?)(?:</arg_value>|(?=<arg_key>)|$)",
-        re.DOTALL,
-    )
-
-    tool_call_start = "<tool_call>"
-    tool_call_end = "</tool_call>"
-
-    def _is_string_type(
-        tool_name: str,
-        arg_name: str,
-        tools: list[Any] | None,
-    ) -> bool:
-        if tools is None:
-            return False
-        for tool in tools:  # pyright: ignore[reportAny]
-            func = tool["function"]  # pyright: ignore[reportAny]
-            if func["name"] == tool_name:
-                params = func["parameters"]  # pyright: ignore[reportAny]
-                if params is None:
-                    return False
-                props = params.get("properties", {})  # pyright: ignore[reportAny]
-                arg_props = props.get(arg_name, {})  # pyright: ignore[reportAny]
-                arg_type = arg_props.get("type", None)  # pyright: ignore[reportAny]
-                return arg_type == "string"  # pyright: ignore[reportAny]
-        return False
-
-    def _deserialize(value: str) -> Any:  # pyright: ignore[reportAny]
-        try:
-            return json.loads(value)  # pyright: ignore[reportAny]
-        except Exception:
-            pass
-        try:
-            return ast.literal_eval(value)  # pyright: ignore[reportAny]
-        except Exception:
-            pass
-        return value
-
-    def parse_tool_call(text: str, tools: list[Any] | None = None):
-        func_name_match = _func_name_regex.search(text)
-        if func_name_match is None:
-            raise ValueError(f"Could not parse function name from tool call: {text!r}")
-        func_name = func_name_match.group(1)
-
-        pairs = _func_arg_regex.findall(text)
-        arg_dct: dict[str, Any] = {}
-        for key, value in pairs:  # pyright: ignore[reportAny]
-            arg_key = key.strip()  # pyright: ignore[reportAny]
-            arg_val = value.strip()  # pyright: ignore[reportAny]
-            if not _is_string_type(func_name, arg_key, tools):  # pyright: ignore[reportAny]
-                arg_val = _deserialize(arg_val)  # pyright: ignore[reportAny]
-            arg_dct[arg_key] = arg_val
-        return dict(name=func_name, arguments=arg_dct)
-
-    tokenizer._tool_call_start = tool_call_start
-    tokenizer._tool_call_end = tool_call_end
-    tokenizer._tool_parser = parse_tool_call
-
-
-def _validate_single_tool(obj: dict[str, Any]) -> ToolCallItem:
-    if (
-        ((name := obj.get("name")) is not None)
-        and ((args := obj.get("arguments")) is not None)
-        and isinstance(name, str)
-    ):
-        raw_id: object = obj.get("id")
-        extra = {"id": str(raw_id)} if raw_id is not None else {}
-        return ToolCallItem(
-            **extra,
-            name=name,
-            arguments=json.dumps(args),
-        )
-    else:
-        raise ValidationError
-
-
 EXO_RUNNER_MUST_FAIL = "EXO RUNNER MUST FAIL"
 EXO_RUNNER_MUST_OOM = "EXO RUNNER MUST OOM"
 EXO_RUNNER_MUST_TIMEOUT = "EXO RUNNER MUST TIMEOUT"
@@ -0,0 +1,72 @@
+import json
+from dataclasses import dataclass
+from typing import Any, Callable
+
+from exo.shared.types.api import ToolCallItem
+
+
+@dataclass
+class ToolParser:
+    start_parsing: str
+    end_parsing: str
+    parse_tool_calls: Callable[[str], list[ToolCallItem] | None]
+
+
+def make_mlx_parser(
+    tool_call_start: str,
+    tool_call_end: str,
+    tool_parser: Callable[[str], dict[str, Any] | list[dict[str, Any]]],
+) -> ToolParser:
+    def parse_tool_calls(text: str) -> list[ToolCallItem] | None:
+        try:
+            text = text.removeprefix(tool_call_start)
+            text = text.removesuffix(tool_call_end)
+            parsed = tool_parser(text)
+            if isinstance(parsed, list):
+                return [ToolCallItem.model_validate(_flatten(p)) for p in parsed]
+            else:
+                return [ToolCallItem.model_validate(_flatten(parsed))]
+
+        except Exception:
+            return None
+
+    return ToolParser(
+        start_parsing=tool_call_start,
+        end_parsing=tool_call_end,
+        parse_tool_calls=parse_tool_calls,
+    )
+
+
+# TODO / example code:
+def _parse_json_calls(text: str) -> list[ToolCallItem] | None:
+    try:
+        text = text.removeprefix("<tool_call>")
+        text = text.removesuffix("</tool_call>")
+        top_level = {
+            k: json.dumps(v) if isinstance(v, (dict, list)) else v
+            for k, v in json.loads(text).items()  # pyright: ignore[reportAny]
+        }
+        return [ToolCallItem.model_validate(top_level)]
+    except Exception:
+        return None
+
+
+def _flatten(p: dict[str, Any]) -> dict[str, str]:
+    return {
+        k: json.dumps(v) if isinstance(v, (dict, list)) else str(v)  # pyright: ignore[reportAny]
+        for k, v in p.items()  # pyright: ignore[reportAny]
+    }
+
+
+json_tool_parser = ToolParser(
+    start_parsing="<tool_call>",
+    end_parsing="</tool_call>",
+    parse_tool_calls=_parse_json_calls,
+)
+
+
+def infer_tool_parser(chat_template: str) -> ToolParser | None:
+    """Attempt to auto-infer a tool parser from the chat template."""
+    if "<tool_call>" in chat_template and "tool_call.name" in chat_template:
+        return json_tool_parser
+    return None
@@ -5,12 +5,13 @@ from typing import Any

 from exo.shared.types.worker.runner_response import GenerationResponse, ToolCallResponse
 from exo.worker.runner.runner import parse_tool_calls
+from exo.worker.runner.tool_parsers import make_mlx_parser


 def _make_responses(
    texts: list[str],
    finish_on_last: bool = True,
-) -> Generator[GenerationResponse | ToolCallResponse]:
+) -> Generator[GenerationResponse]:
    """Create a sequence of GenerationResponses from text strings."""
    for i, text in enumerate(texts):
        is_last = i == len(texts) - 1
@@ -22,10 +23,13 @@ def _make_responses(
        )


-def _dummy_parser(text: str) -> dict[str, Any]:
+def _dummier_parser(text: str) -> dict[str, Any]:
    return {"name": "test_fn", "arguments": {"arg": text}}


+_dummy_parser = make_mlx_parser("<tool_call>", "</tool_call>", _dummier_parser)
+
+
 class TestParseToolCalls:
    """Tests for parse_tool_calls generator."""

@@ -35,8 +39,6 @@ class TestParseToolCalls:
        results = list(
            parse_tool_calls(
                _make_responses(texts, finish_on_last=False),
-                "<tool_call>",
-                "</tool_call>",
                _dummy_parser,
            )
        )
@@ -50,8 +52,6 @@ class TestParseToolCalls:
        results = list(
            parse_tool_calls(
                _make_responses(texts),
-                "<tool_call>",
-                "</tool_call>",
                _dummy_parser,
            )
        )
@@ -76,9 +76,7 @@ class TestParseToolCalls:
        results = list(
            parse_tool_calls(
                _make_responses(texts, finish_on_last=False),
-                "<tool_call>",
-                "</tool_call>",
-                _failing_parser,
+                make_mlx_parser("<tool_call>", "</tool_call>", _failing_parser),
            )
        )
Author	SHA1	Message	Date
Sami Khan	5b81d55d66	lite node	2026-02-19 22:07:34 +05:00
Sami Khan	3e9eb93f82	exo theme	2026-02-19 04:21:58 +05:00
Sami Khan	ab622f79c3	EXO iOS app	2026-02-18 06:40:07 +05:00
Alex Cheema	d6301ed593	dashboard: redesign downloads page as model×node table (#1465 ) ## Motivation The current downloads page uses a node-centric card grid layout that is messy and hard to read — the same model across different nodes appears in separate cards, and deep nesting wastes space. This makes it difficult to quickly see which models are on which nodes. ## Changes Rewrote the downloads page (`dashboard/src/routes/downloads/+page.svelte`) from a card grid to a clean table layout: - Rows = models (unique across all nodes) - Columns = nodes (with disk free shown in header) - Cells show status at a glance: - ✅ Green checkmark + size for completed downloads - 🟡 Yellow percentage + mini progress bar + speed for active downloads - `...` for pending downloads - ❌ Red X for failed downloads - `--` for models not present on a node - Delete/download action buttons appear on row hover - Model name column is sticky on horizontal scroll (for many-node clusters) - Models sorted by number of nodes with completed downloads - Imported shared utilities from `$lib/utils/downloads` instead of inline re-implementations ### Backend: model directory in download events - Added `model_directory` field to `BaseDownloadProgress` so all download status events include the on-disk path - Added `_model_dir()` helper to `DownloadCoordinator` to compute the path from `EXO_MODELS_DIR` - Dashboard uses this to show file location and enable "open in Finder" for completed downloads ### Info modal - Clicking a model name opens an info modal showing card details (family, quantization, capabilities, storage size, layer count, tensor parallelism support) ### Other fixes - Fixed model name truncation in the table - Excluded `tests/start_distributed_test.py` from pytest collection (CLI script that calls `sys.exit()` at import time) ## Test Plan - [x] `uv run basedpyright` — 0 errors - [x] `uv run ruff check` — all passed - [x] `nix fmt` — clean - [x] `uv run pytest` — 188 passed, 1 skipped 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-17 14:31:47 +00:00
Evan Quiney	6d1ca6689b	don't time out node identities (#1493 ) currently nodes leaving and rejoining the cluster can lose their identity. We have no need to delete this data on node timing out, so let's just persist it.	2026-02-17 11:48:28 +00:00
Evan	c01b6fff21	eprint banner our banner was being printed to stdout but should be printed to stderr as its essentially a log message	2026-02-17 11:43:06 +00:00
Jake Hillion	8392e78afe	bench: add spec for automatic canary benchmarks (#1483 ) Adds all the models that can fit onto a single M3 Ultra for single machine benchmarks. Fixes the macOS version, GPU spec, and chip type for maximum reproducibility. Specifies the minimum memory accordingly for each type of model, using the smallest machine available (the smallest M3 Ultra is 96GiB). Test plan: - Running this with some code that makes machines of this spec available and stores the results. It works. This will become part of a larger testing/stability strategy once we've collected more of the data.	2026-02-17 10:52:05 +00:00
Evan	86735ece78	begins begins	2026-02-16 19:26:19 +00:00