rudrankriyam/Foundation-Models-Framework-Lab

Foundation Lab is a native iOS and macOS application designed to serve as an interactive learning and testing environment for Apple's Foundation Models framework. It provides users with a suite of guided experiments, playgrounds, and tools to explore capabilities such as chat, voice interaction, Retrieval-Augmented Generation, and HealthKit integration. The system also includes an Adapter Studio for comparing language model adapters and a command-line interface for training and exporting custom adapters.

25 changes.

This period established the Foundation Lab project, building a native iOS and macOS application for experimenting with Apple's Foundation Models framework. The work focused on implementing core capabilities such as voice-enabled chat, Retrieval-Augmented Generation, and HealthKit integration, while also adding Siri shortcuts and adaptive UI components.

22 changes.

This period focused on expanding the FoundationLabCore library with extensive new capabilities, data models, and provider implementations for diverse AI tasks. It also introduced the Adapter Studio, a macOS-specific tool and CLI for importing, training, and comparing language model adapters, alongside significant UI redesigns for experiments and playgrounds.

Users can now invoke Foundation Lab features directly through Siri using App Intents and App Shortcuts. This change introduces intents for generating book recommendations, estimating meal nutrition, summarizing web pages, and generating localized responses. It also adds intents for interacting with device data and services, including weather, web search, contacts, calendar, reminders, location, music, and health data. A new AppShortcutsProvider registers voice phrases for these actions, enabling hands-free access to the lab's core functionalities.

Foundation Lab/AppIntents · high confidence

A new Python script, Tools/ImageInputProbe/image\_input\_probe.py, has been added to empirically probe Apple Foundation Models for their image-input limits. The tool allows users to test various aspect ratios and image sizes to determine the maximum supported resolution, helping to identify constraints and failure points when submitting images to the model.

python · high confidence

Introduced a set of new Swift structs and enums in the Foundation Lab/AdapterStudio/Models directory to support the Adapter Studio's model comparison feature. These include \AdapterContext\ and \AdapterMetadata\ for adapter information, \ModelCompareEvent\ and \ModelCompareResult\ for tracking comparison progress and outcomes, and \ModelCompareResponseMetrics\ for capturing performance data like time-to-first-token. These models provide the underlying data structures for the UI state (\AdapterStudioColumnState\, \AdapterStudioRunState\) and error handling (\ModelCompareError\) specific to this location.

Foundation Lab/AdapterStudio/Models · high confidence

A new VoiceLogging utility has been introduced to provide structured, categorized logging for voice-related operations. This change adds specific log categories for voice state, recognition, synthesis, permissions, and health data, allowing for more granular debugging and monitoring of voice functionality within the app.

Foundation Lab/Voice · high confidence

The Foundation Lab build process now includes an automated workflow to archive, export, and upload builds to the external TestFlight beta group. This change introduces a new export options template that dynamically populates the Apple Developer Team ID and provisioning profile details, alongside a workflow definition that orchestrates the build pipeline using the \asc\ CLI tool to handle authentication, build number resolution, archiving, and publishing.

.asc · high confidence

A new shell script (Tools/BookPlaygrounds/typecheck.sh) has been added to automatically typecheck Swift source files within the BookPlaygrounds directory. The script builds the project to locate the module path and then runs swiftc with the -typecheck flag on all .swift files found in subdirectories, ensuring that sample code remains valid against the built modules.

Tools/BookPlaygrounds · high confidence

The Foundation Lab view models now support a suite of new interactive experiments, including a Dynamic Profile workflow for inspecting and reviewing model behavior, a Gemini video input lab for analyzing video content, an image input lab for alt-text and vision tasks, a reasoning level comparison tool for Private Cloud Compute, a Spotlight RAG lab for searching local device data, and a tool-calling mode comparison lab. Additionally, the ChatViewModel now includes full voice interaction support with speech recognition and synthesis, and the RAGChatViewModel provides a document Q&A interface with local indexing.

Foundation Lab/ViewModels · high confidence

Foundation Lab is now available as a standalone application, bringing a native macOS experience with dedicated window management, keyboard shortcuts for navigation (Library, Playground, Runs), and a Settings view. The app integrates HealthKit and Private Cloud Compute capabilities, allowing users to explore dynamic profiles and health-related AI workflows. It features a redesigned chat interface with voice input support, token counting in results, and multi-language localization for key UI elements and app shortcuts.

Foundation Lab · high confidence

This change establishes the initial structure for the Foundation Lab project, a native iOS and macOS tool for learning and testing Apple's Foundation Models framework. It introduces core documentation files (README, Agents, CLAUDE, PRODUCT) that define the app's purpose, architecture, and design principles, alongside configuration files for SwiftLint and Xcode build settings. The repository also includes a detailed API delta document (fmf.md) tracking changes in the Xcode 27 Foundation Models SDK and a .gitignore file to manage build artifacts and local configurations.

(repo-wide) · high confidence

Foundation Lab now includes Adapter Studio, a new service and UI layer that allows users to import, manage, and compare \.fmadapter\ packages against the base language model. Users can import adapter packages via an open panel, which are then stored in the Application Support directory, and run side-by-side generation comparisons to evaluate adapter performance. The feature includes a dedicated view model to manage run states, handle streaming responses, and display comparison results, along with robust error handling for invalid files, directory permissions, and large file sizes.

Foundation Lab/AdapterStudio/Services · high confidence

The Foundation Lab/Models area now provides the core data models and persistence layer for the app's experiment library. This includes the \ExperimentLibraryRepository\ and \ExperimentStore\ for managing saved experiments and run history, backed by a new \ExperimentPersistenceDocument\ and \ExperimentPersistenceRepository\ that handle JSON storage with schema versioning and legacy migration. The \ExperimentTemplate\ and \ExampleType\ enums define the curated catalog of guided labs, recipes, and workshops, while \AppConfiguration\ centralizes settings like token management and health session timeouts. Additionally, new models such as \ChatGenerationOutcome\ and \ChatMessage\ support the chat interface, and \DefaultPrompts\ provides sample content for various example types.

Foundation Lab/Models · high confidence

The Foundation Lab view introduces a new chat interface featuring voice input capabilities, allowing users to start voice mode, listen to partial transcripts, and send voice messages. The chat input dynamically switches between text entry and voice controls, while the message display includes a typing indicator and context summary labels. Additionally, the lab now tracks and displays token usage via a progress bar that visualizes context window consumption with color-coded warnings, and result displays show token counts alongside copy functionality.

Foundation Lab/Views/Components · high confidence

The \fmas\ command-line interface is now available in \Tools/AdapterStudio/adapter\_cli\, providing a unified entry point for Foundation Models adapter workflows. Users can now initialize and validate their toolkit environment (\fmas init\), set up the Python virtual environment and dependencies (\fmas setup\), and execute core operations including text generation with base or trained models (\fmas generate\), adapter training (\fmas train-adapter\), draft model training for speculative decoding (\fmas train-draft\), and exporting trained adapters to the \.fmadapter\ format (\fmas export\). The CLI includes robust input validation, error handling, and state preservation to ensure reliable interaction with the underlying training toolkit.

_Tools/AdapterStudio/adapter\cli · high confidence

Users can now interact with a new Health Chat view that displays a token usage bar at the top and provides quick-action chips (e.g., for steps, sleep, and weekly summaries) to easily query Health data. The interface includes a text input field, message bubbles for user and system responses, and support for copying messages, while automatically scrolling to new content and handling conversation deletion.

Foundation Lab/Health/Views/Chat · high confidence

A new HealthKitService actor has been added to handle HealthKit integration, including requesting read authorization for steps, active energy, distance, heart rate, and sleep data, and fetching daily and weekly health metrics.

Foundation Lab/Health/Services · high confidence

Foundation Lab now includes a new Adapter Studio workspace (macOS only) that lets users import .fmadapter packages and run side-by-side comparisons of a base system model against a custom adapter. The studio provides a multi-stage interface: Settings for importing and managing saved adapters, Runs for entering prompts and executing live comparisons, Evaluation for viewing timing metrics (time to first token, total duration, and deltas), Output for reviewing generated responses, and Preview for workflow guidance. On non-macOS platforms, the view displays an unavailability message directing users to the fmas CLI for training and export.

Foundation Lab/AdapterStudio/Views · high confidence

Added a new HealthRepository class to handle the persistence of health metrics using SwiftData. This component manages the insertion and retrieval of health data, ensuring that metrics are saved with consistent timestamps and fetched based on configurable date ranges, while operating on the main actor to comply with SwiftData's thread-safety requirements.

Foundation Lab/Health/Repositories · high confidence

The DynamicSchemas example area now includes a comprehensive set of SwiftUI views and helper utilities demonstrating runtime schema generation. New views cover basic object schemas (BasicDynamicSchemaView), arrays with min/max constraints (ArrayDynamicSchemaView), enum-based classifications (EnumDynamicSchemaView), guided patterns with constraints like regex and ranges (GuidedDynamicSchemaView), compile-time @Generable types (GenerablePatternView), and a form builder that constructs schemas from natural language descriptions (FormBuilderSchemaView). These views are supported by shared helper files (DynamicSchemaHelpers, ArrayDynamicSchemaHelpers, EnumDynamicSchemaHelpers, FormBuilderSchemaHelpers, GenerablePatternHelpers, GuidedDynamicSchemaHelpers) and a new DynamicSchemaExecutorExtension that provides convenience methods for executing dynamic schema generations and formatting results as JSON.

Foundation Lab/Views/Examples/DynamicSchemas · high confidence

This change introduces the foundational data models for the redesigned Foundation Lab experiments. It adds new types to define experiment configurations (including runtime, reasoning level, and tool selection), experiment runs (capturing prompts, responses, token usage, and status), and experiment events (tracking user, assistant, and tool interactions). It also defines the built-in tools available for experiments (such as weather, web search, contacts, calendar, reminders, location, health, music, and web metadata), including specific logic for requiring confirmation on calendar and reminder mutations. Additionally, it adds domain-specific models for capabilities like book recommendations, nutrition analysis, journaling, product reviews, and story outlining, along with a tool trajectory evaluation model for comparing expected versus observed tool calls.

FoundationLabCore/Sources/FoundationLabCore/Models · high confidence

Foundation Lab now includes dedicated example views for structured data generation, generation guides, and Retrieval-Augmented Generation (RAG). The StructuredDataView demonstrates generating book recommendations conforming to a Swift type, while the GenerationGuidesView shows how to constrain responses using @Guide annotations. A new RAGChatView enables document Q&A with source citation, and the ExampleType+Destination extension wires these new examples into the app's navigation alongside existing ones like ModelAvailability and Health.

Foundation Lab/Views/Examples · high confidence

Added two new tools to the Foundation Lab core: \readFoundationLabFact\, which provides a deterministic, read-only way to access fixed in-memory facts about transcripts, tools, and privacy, and \searchWeb\, which enables web searches via the Search1API keyless endpoint. These additions allow labs to observe session details without external dependencies and to perform live web lookups directly within the model interaction.

FoundationLabCore/Sources/FoundationLabCore/Tools · high confidence

This change introduces a suite of new use cases and result types in FoundationLabCore/Capabilities, enabling users to analyze nutrition, generate book recommendations, create health encouragement messages, and run dynamic schema examples (including basic objects, arrays, and enums). It also adds capabilities for multilingual responses, web page summarization, location, weather, reminders, calendar, health data, contacts, music catalog, and web search, along with a language session demo. These additions expand the core library's ability to orchestrate Foundation Models for diverse, structured, and conversational tasks.

FoundationLabCore/Sources/FoundationLabCore/Capabilities · high confidence

The FoundationLabCore library now exposes a set of new public protocols and corresponding Foundation Models implementations for capabilities including book recommendations, calendar querying, contacts searching, conversation running, health data querying, health encouragement generation, location responding, music catalog searching, nutrition analysis, reminder management, weather responding, web page summarization, and web searching. These providers enable the app to leverage Foundation Models for these specific tasks, with implementations handling tool invocation, prompt construction, and result formatting.

FoundationLabCore/Sources/FoundationLabCore/Providers · high confidence

FoundationLabCore now includes a comprehensive set of new request structs in the Requests directory, enabling developers to invoke specific AI capabilities such as nutrition analysis, book recommendations, health encouragement, web page summarization, weather lookups, location services, reminder management, calendar queries, health data queries, conversation running, language demos, schema examples, and contact/music/web searches. These structs conform to FoundationModelCapabilityRequest and integrate with FoundationModelsKit, providing a standardized interface for tool-backed features within the Foundation Lab ecosystem.

FoundationLabCore/Sources/FoundationLabCore/Requests · high confidence

A new Health dashboard view has been added to the Foundation Lab, displaying today's HealthKit metrics including steps, heart rate, sleep, active energy, and distance. The interface handles loading states, empty data scenarios, and authorization errors, providing a 'Try Again' option if data is unavailable. Users can also initiate a health-focused chat via a new 'Ask About Health Data' button in the toolbar, which presents a HealthChatView sheet.

Foundation Lab/Health/Views/Dashboard · high confidence

A new \fetchHealthData\ tool has been added to the Foundation Lab, allowing AI agents to retrieve authorized HealthKit measurements. Users can now request specific metrics such as steps, heart rate, sleep, active energy, and distance, either for the current day, the past week, or individual metrics. The tool supports refreshing data directly from HealthKit or reading from a local cache, and provides structured JSON responses indicating the status (success, partial, or unavailable) of the requested health data.

Foundation Lab/Health/Tools · high confidence

The Foundation Lab now includes new interactive examples for Xcode 27, featuring a Context Budget Visualizer to simulate token usage and policy decisions, a Context Window Inspector for real-time token measurement, an Agent Flow Inspector to map framework boundaries, and a Dynamic Profile Builder to demonstrate live session profile changes.

Foundation Lab/Views/Examples/Xcode27 · high confidence

The chat interface in Foundation Lab now includes a new RAG document picker that lets users import local files (PDF, Markdown, plain text, HTML, RTF) or load sample sources to enable grounded, document-based Q&A. The updated transcript view displays individual message entries with real-time token usage counts and, on iOS 27+, shows expandable reasoning traces when available. Tool calls and their results are presented in expandable disclosure groups for better transparency, and a lightweight activity indicator view supports status feedback during operations.

Foundation Lab/Views/Chat · high confidence

Added three new catalog files to FoundationLabCore that provide structured example data for the application's demo and learning features. FoundationLabExampleCatalog defines nine demo scenarios (such as basic chat, streaming, and journaling) with default prompts and suggestions. FoundationLabLanguageCatalog introduces multilingual support with conversation steps and prompt templates for various languages. FoundationLabSchemaCatalog provides schema presets for generating structured data, including basic objects, arrays, and enumerations with specific use-case examples.

FoundationLabCore/Sources/FoundationLabCore/Catalogs · high confidence

The app introduces a new data architecture for health features, adding \HealthDataManager\ to coordinate health metrics and authorization, \HealthMetric\ to model and format specific health data points (such as steps, heart rate, and sleep), and \HealthSession\ to track chat interactions and session types. This change establishes the foundational models and persistence logic required for the health chat and analysis capabilities.

Foundation Lab/Health/Models · high confidence

This update adds a comprehensive set of Swift Playgrounds in the BookPlaygrounds directory that demonstrate how to use the FoundationModels framework. The new examples cover session management (availability checking, single and multi-turn conversations, context windows, and prewarming), generation control (temperature, token limits, sampling strategies like greedy and top-K/P, and output constraints), and tool use (implementing the Tool protocol for search, calculation, weather, and location services, including multi-tool coordination and error handling). These playgrounds serve as practical, runnable references for integrating and configuring Foundation Models in applications.

BookPlaygrounds · high confidence

The Foundation Lab now includes dedicated views for exploring language capabilities: Language DetectionView lists supported languages and locales reported by the runtime; MultilingualResponsesView runs the same task across multiple languages to compare model outputs; SessionManagementView demonstrates maintaining context while switching languages within a single conversation; and ProductionLanguageExampleView shows how to adapt structured model responses (such as nutrition analysis) to the user's selected language. These views are wired into the app's navigation via LanguageExample+Destination and rely on LanguageService and FoundationModelsKit runtime APIs.

Foundation Lab/Views/Languages · high confidence

The Foundation Lab example views now use a new set of base components to standardize the user experience. ExampleExecutor provides a unified interface for running basic text generation, structured data generation, and book recommendation tasks, including token usage tracking and centralized error handling. ExampleViewBase and ReferenceExampleView offer consistent layouts for interactive and documentation-only examples, featuring standardized prompt inputs, run/reset controls, and error display. CodeViewer adds syntax-highlighted code blocks with copy functionality, while PromptHistory allows users to quickly select from recent prompts.

Foundation Lab/Views/Examples/Components · high confidence

The Runs section now features a dedicated list view that groups and searches experiment runs, allowing users to delete individual runs or all runs at once. Tapping a run opens a detail view that displays the full transcript, configuration (including model, runtime, and generation options), and status, with an option to load the configuration into the Playground for reuse.

Foundation Lab/Views/Runs · high confidence

Foundation Lab introduces three new service classes to support its redesigned experiment architecture. LanguageService manages supported languages and displays user locale information. RAGService handles document indexing, text chunking, semantic search, and database reset operations using LumoKit and VecturaKit, including safe copying of imported documents to app storage. GeminiDeveloperVideoLanguageModel provides a new LanguageModel implementation for video-based interactions with the Gemini API, restricted to text responses and excluding tool calling capabilities.

Foundation Lab/Services · high confidence

The Health Chat feature now uses a dedicated HealthChatViewModel that integrates with FoundationModelsKit to provide accurate, real-time token usage tracking via the SystemLanguageModel.TokenUsage API, replacing previous synthetic or fallback counting methods. This change ensures users see precise context window utilization and token consumption metrics during health data conversations, improving transparency and helping manage session limits effectively.

Foundation Lab/Health/ViewModels · high confidence

The app now features a new AdaptiveNavigationView that automatically switches between a tab-based interface on compact screens (iPhone) and a split-view layout on larger devices (iPad/macOS), persisting sidebar visibility and selection state. A new WorkspaceView provides a structured, phase-based interface for experiments like Adapter Comparison, complete with a stage picker, phase headers, and a status rail. Additionally, a ModelUnavailableView offers specific guidance and settings access when Apple Intelligence is not available, and a SettingsView displays app version and support links.

Foundation Lab/Views · high confidence

A new Color extension has been added to define the app's primary accent color as blue and establish platform-adaptive secondary and tertiary background colors for iOS and macOS, ensuring a consistent visual style across views and components.

Foundation Lab/Extensions · high confidence

The Playground has been redesigned around the concept of 'Experiments', allowing users to configure and save distinct model setups (including runtime, reasoning level, and generation options) via a new inspector. A key behavioral change is the introduction of a mandatory approval workflow for destructive tool mutations (such as calendar or reminders writes); users must now explicitly approve these actions via a dedicated sheet before they are executed, replacing the previous implicit or unconfirmed behavior.

Foundation Lab/Views/Playground · high confidence

The FoundationLabCore module now exposes a new \FoundationLabCoreError\ enum that provides structured, localized error messages for common failure scenarios. Users will see specific, user-friendly descriptions for invalid requests, unavailable capabilities, provider failures, and unsupported environments, replacing any previous generic error handling with these distinct, localized cases.

FoundationLabCore/Sources/FoundationLabCore/Errors · high confidence

The Library interface has been redesigned to organize content into 'My Experiments' (saved user configurations) and curated 'Recipes' (templates). Users can now search across both saved experiments and library templates, and recipes are launched as editable Playground sessions rather than static examples. The view also introduces a new catalog system for Xcode 27 features and supports deletion of saved experiments via swipe actions.

Foundation Lab/Views/Library · high confidence

The voice functionality in Foundation Lab has been refactored to improve stability and user experience. The new PermissionManager service centralizes microphone and speech recognition permission requests, preventing looping permission prompts and providing clearer feedback when access is denied. Speech recognition and synthesis services have been updated to use modern Swift concurrency patterns, including AsyncStream for state updates and proper actor isolation to prevent main-thread blocking and crashes. Audio buffer handling has been hardened to prevent corruption during streaming, and task cancellation is now properly supported to ensure clean shutdowns when voice interactions are interrupted.

Foundation Lab/Voice/Services · high confidence

Added test coverage for FoundationLabCore capabilities and infrastructure; Added tests for the Image Input Probe tool; Added unit tests for Adapter Studio CLI, configuration, and validation.

The Foundation Lab Xcode project has been configured to include the HighlightSwift, FoundationModelsKit, FoundationLabCore, and LumoKit packages as framework dependencies. The project build settings reference a SampleCode.xcconfig file, and the application bundle is set to use the 'Foundation Lab.icon' resource.

FoundationLab.xcodeproj · high confidence

The project introduces its initial dependency structure, pinning the Swift package \FoundationModelsKit\ (revision 65834b11) and its \FoundationModelsTools\ component for the \FoundationLabCore\ library, while also establishing a new \FoundationLabCore\ package target. The Swift toolchain is configured for iOS 26, macOS 26, and visionOS 26, and includes additional pinned dependencies such as \HighlightSwift\ 1.1.0, \LRUCache\ 1.2.0, \PicoDocs\ 1.0.1, \swift-argument-parser\ 1.7.0, \swift-collections\ 1.3.0, \swift-jinja\ 2.2.0, \swift-safetensors\ 0.0.8, \swift-sentencepiece\ 0.0.6, \swift-transformers\ 1.1.6, and \SwiftSoup\ 2.11.3. A Python-based CLI tool, \foundation-models-adapter-studio\ (version 0.2.0), is also added with a dependency on \setuptools\ and \wheel\.

(dependencies) · high confidence

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

• CAI 67 → 64 (-3.3)
• Rubric changed (rubric-2026.09.11 → rubric-2026.09.18) — scores are not directly comparable.

• Code Health 92 → 91 (-1.1)
• Architecture 99 → 98 (-0.9)
• Maturity 74 → 74 (-0.1)
• Readiness 55 → 48 (-6.9)
• Security 70 → 70 (+0.0)
• Domain Modelling 100 → 100 (+0.0)

• Dependency hygiene PARTLY measured — SwiftPM pinning read, dependency currency NOT established
• Documentation: no installation or build instructions (README.md)

• Coverage not measured — Swift suite
• Duplicated block (10 lines × 2) (Foundation Lab/Views/AdaptiveNavigationView.swift)
• Duplicated block (10 lines × 2) (Foundation Lab/Views/Examples/Xcode27/FMCLIPythonPlaygroundView.swift)
• Duplicated block (10 lines × 3) (Foundation Lab/Views/Examples/DynamicSchemas/ArrayDynamicSchemaView.swift)
• Duplicated block (10 lines × 4) (Foundation Lab/Views/Examples/DynamicSchemas/GenerablePatternView.swift)
• Duplicated block (11 lines × 2) (Foundation Lab/Views/Examples/Xcode27/ReasoningComparisonResultView.swift)
• Duplicated block (11 lines × 3) (Foundation Lab/Views/Examples/Xcode27/DynamicProfileBuilderView.swift)
• Duplicated block (12 lines × 2) (Foundation Lab/Health/Views/Chat/HealthChatView.swift)
• Duplicated block (12 lines × 2) (Foundation Lab/ViewModels/DynamicProfileWorkflowViewModel.swift)
• Duplicated block (12 lines × 2) (Foundation Lab/Views/Examples/DynamicSchemas/BasicDynamicSchemaView.swift)
• Duplicated block (12 lines × 3) (Foundation Lab/Health/Models/HealthMetric.swift)
• Duplicated block (12 lines × 4) (Foundation Lab/Views/Examples/Xcode27/DynamicProfileBuilderLiveView.swift)
• Duplicated block (13 lines × 2) (Foundation Lab/AdapterStudio/Views/AdapterStudioEvaluationView.swift)
• Duplicated block (13 lines × 2) (Foundation Lab/Views/Examples/Xcode27/ReasoningLevelComparisonLiveView.swift)
• Duplicated block (13 lines × 2) (Tools/AdapterStudio/adapter_cli/commands/setup.py)
• Duplicated block (13 lines × 4) (Foundation Lab/Views/Examples/Components/ExampleViewBase.swift)
• Duplicated block (13 lines × 5) (Tools/AdapterStudio/adapter_cli/commands/demo.py)
• Duplicated block (14 lines × 2) (Foundation Lab/Views/Examples/Xcode27/DynamicProfileBuilderLiveView.swift)
• Duplicated block (14 lines × 2) (Foundation Lab/Views/Examples/Xcode27/ToolCallTrajectoryViewerView.swift)
• Duplicated block (14 lines × 3) (Foundation Lab/Views/Examples/DynamicSchemas/ArrayDynamicSchemaView.swift)
• …and 59 more

• 2 commits — 2 feature/other, 0 fixes

• (repo) — 1 commit
• (root) — 1 commit

• change: Merge pull request #217 from rudrankriyam/codex/update-fmf-xcode-27-2-beta
• change: docs: update Foundation Models API for Xcode 27.2 beta

• Unchanged — 0 containers · 1 contexts · 0 edges

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

rudrankriyam/Foundation-Models-Framework-Lab was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.