v2.23.5 New features: - Generator LLM (MLX): New on-device LLM/VLM inference generator using MLX framework - Currently only available with Buttress, native support coming soon Bug fixes: - Generator LLM (GGML): Fix SIGABRT crash during model loading with Metal - Standby: Restore rendering stability and debounce Subspace change-canvas updates - Prevent unhandled promise rejection from auth errors - Other stability improvements and bug fixes

v2.23.4 - System->Take Screenshot action: Capture full viewport including portal Subspaces - UDP module stability improvements - Reduce CPU usage significantly - Standby Transition & Animation: Fix rendering, breathe animation, and scaling on canvas switch and resize - Clear image memory cache on memory warning to prevent main thread hangs - Fix EXC_BAD_ACCESS crash on app exit caused by System Data Expression - Generator LLM (GGML): Bump llama.cpp runtime to b8195 - Suppress unhandled network errors in Apollo GQL error handlers - Prevent crash when channel publishes to unmounted Subspace - Other stability improvements and bug fixes

v2.23.3 Bug fixes: - Fix crashes on app termination (Metal context, worklets destructor) - Preserve menu and Marked Apps shortcut events - Generator MQTT: Additional fix for SIGSEGV crash during reconnection - Generator HTTP: Parse SSE error event data for meaningful error messages - Brick Items: Preserve standby transition with post transform script - DevTools: Stabilize server startup and restart lifecycle - Generator HTTP Server: Guard socket writes against closed sockets - Data Bank Runtime: Improve memory management - Other stability improvements and bug fixes

v2.23.2 Features: - Generator VAD (ONNX) & VAD (Traditional): New voice activity detection generators for real-time speech detection - Generator STT (GGML): Add TRANSCRIBE_STOP action - Generator LLM (GGML): Bump llama.cpp runtime to b8095 - DevTools: Add UDP LAN discovery for automatic server scanning - Marked Apps: Mark/favorite applications for quick binding (up to 5 per workspace) Bug fixes: - Generator MQTT: Fix multiple SIGSEGV crashes during reconnection - Fix RetryLink to skip non-retriable errors (403, timeout, abort) across all GraphQL clients - Fix various runtime crashes with graceful error handling - Other stability improvements and bug fixes

v2.23 Highlights: - AI/ML Features Now Stable: The Assistant, MCP, and MCP Server generators are now production-ready, enabling AI-powered content generation and Model Context Protocol integration in your applications - BRICKS Buttress (Experimental): Offload compute-intensive AI tasks to external servers. Includes LLM and STT backends with auto-discovery and smart fallback - Apple Intelligence (iOS 26+): Native integration with Apple's on-device AI - LLM with streaming, speech-to-text, and high-quality speech synthesis - Brick Maps (Pro+): Comprehensive mapping with Google Maps and Apple Maps support, markers, path drawing, custom styling, and full interactivity - Data Calculation Trigger Mode: Choose between Auto mode (triggers on input change) and Manual mode (explicit trigger only) - Performance Improvements: Faster data operations on separate thread, optional chaining in expressions, Skia renderer for charts Breaking Changes: - Camera Brick: Zoom now uses actual zoom factor (1x, 2x) instead of 0-1 range. autoFocusEnabled and focusDepth properties removed - Brick 3D Viewer has been removed

v2.22.10 BRICKS Foundation v2.22.10 features & bug fixes - Brick Image: Add templateData/templateType properties for dynamic URL interpolation - Data Operation: Add built-in globals and IIFE variable bindings for expressions - Generator MCP Server: Fix TypeError in clientError handler - DevTools: Fix HTTP request body buffering for large payloads - Fix race condition issues in Subspace and property bank lock - Fix 403 auth errors causing unnecessary retry requests - Fix System->OPEN_FILE action to support array file types - Other stability improvements and bug fixes

v2.22.9 BRICKS Foundation v2.22.9 features & bug fixes - Debug Panel: Add Servers tab showing running MCP/HTTP/TCP servers with connection info - Brick WebView: Add scroll control properties (scrollEnabled, nestedScrollEnabled) - Generator HTTP: Add response selector property to extract specific data from responses - Generator HTTP Server, TCP Server: Add IS_LISTENING Outlet - Generator MCP: Add script member support for programmatic control - Generator Assistant: Add MCP tool blocking, llmToolCallFallbackContent option, finish reason in events - Generator LLM (GGML): Bump llama.cpp runtime to b7617 - Other stability improvements and bug fixes

v2.22.8 BRICKS Foundation v2.22.8 features & bug fixes - Brick Generative Media: Add Veo 3.1 video generation with resolution, duration, and person generation settings - Generator Assistant: Add ON_HEATUP_FINISHED, ON_GENERATING, ON_FINISHED events; support MCP resource annotations - Generator HTTP: Support arguments for HTTP request action and script member - Generator HTTP Server, MCP Server: Normalize path to handle trailing slashes - Generator LLM (GGML): Bump llama.cpp runtime to b7493 - Add TOON encode/decode commands to JS sandbox and string commands - Fix client crash when queries are in flight during logout - Other stability improvements and bug fixes

v2.22.7 BRICKS Foundation v2.22.7 features & bug fixes - Brick Text Input: Fix outlet/event not triggering on clear() - Brick Generative Media: Add gemini-3-pro-image-preview model support - Generator Serial Port: Catch USB DETACH error gracefully - Generator MQTT: Fix crash from nil message data on iOS - Generator Data Bank: Add client guard in subscribeDataChange to prevent errors - Generator STT (GGML): Correct model quants and remove non-existent large-v3-q8_0 model - Generator VAD (GGML): Change default model to silero-v6.2.0 - Generator LLM (GGML): Bump llama.cpp runtime to b7342, support Clear Cache action - Remote Debug: Improve error handling for connection close and Firebase permission errors - Handle persisted cache restore failures in edge cases - Other stability improvements and bug fixes

v2.22.6 BRICKS Foundation v2.22.6 features & bug fixes - Debug Panel: Add copy value button - Generator Question: Support HID keyboard based scanner with flush mode options - Generator LLM (GGML): Bump llama.cpp runtime to b7205 - Other stability improvements and bug fixes

v2.22.5 BRICKS Foundation v2.22.5 features & bug fixes - Generator Assistant: Avoid crash on non-iterable initialMessages - Generator Generative Media: Add Gemini API support - Generator LLM (GGML): - Bump llama.cpp runtime to b7082 - Improve performance of CPU backend - Add OpenCC support to JS Sandbox - Other stability improvements and bug fixes

v2.22.4 BRICKS Foundation v2.22.4 features & bug fixes - Generator SQLite: Support transactions, batch operations, and metadata queries - Generator Tick: Support running state outlet - Generator Iterator: Support FIRST action and FIRST/END event triggers - Generator Assistant: Fix wrong session ID for files outlet - Brick Generative Media: Fix Lottie animations not working - Generator LLM (GGML): Bump llama.cpp runtime to b6870 - Improve native video player resource cleanup - Other stability improvements and bug fixes

v2.22.3 BRICKS Foundation v2.22.3 features & bug fixes - Generator LLM (GGML) - Convert regex shortcuts to GBNF-compatible patterns for tools - Bump llama.cpp runtime to b6816 - Generator Canvas Map: Enhance with outlets, events, and improved type safety - Generator Generative Media: Improve API usage & add loading outlet - Generator AlarmClock: Fix critical bugs and improve code quality - Generator Assistant: Improve vector store usage with proper URL file path handling and error catching - Generator Web Crawler: Improve code quality and fix anti-patterns - Activity Log: Add instant upload trigger for activity logs - Application Filter: Add Claude Sonnet 4.5 and Haiku 4.5 models - System Data: Add GGML Backend Devices - Other stability improvements and bug fixes

v2.22.2 BRICKS Foundation v2.22.2 features & bug fixes - Generator HTTP Server - Correct CORS implementation and improve routes management - Add heartbeat to avoid text/event-stream idle timeout - Generator LLM (GGML): Bump llama.cpp runtime to b6730 - Generator LLM (OpenAI-compat, Anthropic-compat): Fix LLM result key handling - Generator Serial Port: Improve data handling for undefined/null values - Generator STT (GGML): Improve stability - System Data: Add periodic WiFi network details updates - Launch App: Preload styles optimization - Improve screenshot error handling - Fix file type check in open file operations - Other stability improvements and bug fixes

v2.22.1 BRICKS Foundation v2.22.1 features & bug fixes - Generator Data Bank: Fix propertyId handling and remove unnecessary close() call - Generator LLM: Bump llama.cpp runtime to b6569 - Improve error handling and logging in launcher - Improve i18n to respect user's preferred language order - Add user-agent header for file load requests - Add base headers support for URL file synchronization

v2.22.0 BRICKS Foundation v2.22.0 features & bug fixes - Support Application Filter (Preview) - Support `Long Press` event for pressable bricks - Generator MCP Server - Support `script` handler type - Support `Response Type` property for response/tool/prompt - Generator HTTP Server - Support `Cors Origins` and `Res Header` properties - Support TLS properties - Generator TTS (ONNX): Fix error when using `bert-vits2` model - Generator Realtime Transcription - Support `Begin Transcribe Script` and `Begin Transcribe Script Variables` properties for script to determine whether to proceed with transcription - Generator Assistant: Improve stability - Preload - New UI update for better display - Check file integrity - Animation & Standby Transition: Support custom easing function - Local Debugging (Preview) - Logs now pipe to local file instead of in-memory - Improve logs download performance - MCP: Add new tools: - `searchMediaFiles`: Search media files from current used Media boxes of application - `exportSnapshots`: Export updated application state by tool like `updateBrick`, or Application Filter - `importSnapshots`: Import snapshots to update application state - Advanced Setting: Add AI API key fields (OpenAI/Anthropic/Gemini) - Support new System Data - OpenAI API Key from Advanced Setting - Anthropic API Key from Advanced Setting - Gemini API Key from Advanced Setting - AI Filters - NOTE: API Keys only available if application setting `Enable AI System Data` is enabled - Support new System Actions - TRIGGER_APPLICATION_FILTER - Trigger application AI filter - STOP_APPLICATION_FILTER - Stop running application AI filter - SAVE_UPDATE_SNAPSHOT - Save update snapshot (from Application Filter or DevTools MCP Server) to media system - Other bug fixes

v2.21 New Features and Changes: - Add local debugging via Chrome DevTools Protocol. Open `Advanced Settings` to enable - Add new System Action: Data Expression - LLM context limit now defaults to 99 - New Brick: Rive (Preview) - Support for Rive animations - Brick Image: Improve performance and rendering with new image engine - Generator Question: Support more input types and improved UI - New types: language, numeric-string, preference, voice-record, video-record, cart, qty-list - Improved UI for yes/no questions and code scanning - Generator File: Support set content-type and fields in downloadFile - Fix device ID issues when restoring from backup Breaking Changes: - Remove Generator TensorFlowInference - Event handling is now asynchronous and can be configured to run sequentially - Remove gpt3-tokenizer from JS Sandbox - Add locking mechanism in calculations (Disable it in Application Settings when editing) v2.21.14 - System: Prevent locked devices from rebinding via System Menu - Generator LLM (GGML) - Bump llama.cpp runtime to b6327 - Support completionReasoningFormat and useFlashAttn enum migration - Generator Assistant: Enhanced submit & cancel mechanism with pre-fill assistant message support - Generator MCP Server: Support for `static` tool functionality - Generator Realtime Transcription: Event improvements and enhanced functionality

v2.21 New Features and Changes: - Add local debugging via Chrome DevTools Protocol. Open `Advanced Settings` to enable - Add new System Action: Data Expression - LLM context limit now defaults to 99 - New Brick: Rive (Preview) - Support for Rive animations - Brick Image: Improve performance and rendering with new image engine - Generator Question: Support more input types and improved UI - New types: language, numeric-string, preference, voice-record, video-record, cart, qty-list - Improved UI for yes/no questions and code scanning - Generator File: Support set content-type and fields in downloadFile - Fix device ID issues when restoring from backup Breaking Changes: - Remove Generator TensorFlowInference - Event handling is now asynchronous and can be configured to run sequentially - Remove gpt3-tokenizer from JS Sandbox - Add locking mechanism in calculations (Disable it in Application Settings when editing) v2.21.13 - New Generator: LLM (Anthropic-compat) (Preview). Supported by Generator Assistant - Generator MCP / Assistant: Support register global generator with key (Global Generator Key) - Local Debugging (Preview): MCP: Use Data Manager for runtime props retrieval - System: PROPERTY_BANK_EXPRESSION validates `variables` is an array - Generator LLM (GGML): Fix “Use KV Unified” setting not applied

v2.21 New Features and Changes: - Add local debugging via Chrome DevTools Protocol. Open `Advanced Settings` to enable - Add new System Action: Data Expression - LLM context limit now defaults to 99 - New Brick: Rive (Preview) - Support for Rive animations - Brick Image: Improve performance and rendering with new image engine - Generator Question: Support more input types and improved UI - New types: language, numeric-string, preference, voice-record, video-record, cart, qty-list - Improved UI for yes/no questions and code scanning - Generator File: Support set content-type and fields in downloadFile - Fix device ID issues when restoring from backup Breaking Changes: - Remove Generator TensorFlowInference - Event handling is now asynchronous and can be configured to run sequentially - Remove gpt3-tokenizer from JS Sandbox - Add locking mechanism in calculations (Disable it in Application Settings when editing) v2.21.12 - Add workspace name in waiting screen - Support i18n for menu item double-check - Brick Items: Improve stability - Generator LLM (GGML): Bump llama.cpp runtime to b6018 and support "Use SWA Full" property - Generator TTS (ONNX / OpenAI-like / GGML): Fix player not release after application finalize - Local Debugging (Preview): - Home Page: Add check passcode button and click URL to copy functionality - CDP: Support copy object, add globals for inspecting Generator MCP Server, add memoryUsage and diskUsage globals for system monitoring - MCP: Split getBricks & updateBricks tools, add includeStaticProperties parameter, rename properties key for clarity, update data link passthrough, improve tool descriptions, rename props to avoid JSON schema confusion, improve prompt for switch functionality, use small batch size in llama embedding - Improve handling for download logs

v2.21 New Features and Changes: - Add local debugging via Chrome DevTools Protocol. Open `Advanced Settings` to enable - Add new System Action: Data Expression - LLM context limit now defaults to 99 - New Brick: Rive (Preview) - Support for Rive animations - Brick Image: Improve performance and rendering with new image engine - Generator Question: Support more input types and improved UI - New types: language, numeric-string, preference, voice-record, video-record, cart, qty-list - Improved UI for yes/no questions and code scanning - Generator File: Support set content-type and fields in downloadFile - Fix device ID issues when restoring from backup Breaking Changes: - Remove Generator TensorFlowInference - Event handling is now asynchronous and can be configured to run sequentially - Remove gpt3-tokenizer from JS Sandbox - Add locking mechanism in calculations (Disable it in Application Settings when editing) v2.21.11 - Canvas Control: Fix schedule cannot detect devices with APP_CONTROL_GROUP - TTS (GGML & ONNX) - Speaker: Support create OuteTTS speaker config - Local Debugging (Preview): - Implement auto port detection for server - MCP: Brick supports switch props - Generator LLM (GGML): Improve tool_calls format and use content instead of text - Generator TTS (GGML & ONNX): Support change ubatch size for LLM part, batchSize for vocoder model, and add completionPrompt property - Generator Assistant (Preview): Put content for better response on tool_calls

v2.21 New Features and Changes: - Add local debugging via Chrome DevTools Protocol. Open `Advanced Settings` to enable - Add new System Action: Data Expression - LLM context limit now defaults to 99 - New Brick: Rive (Preview) - Support for Rive animations - Brick Image: Improve performance and rendering with new image engine - Generator Question: Support more input types and improved UI - New types: language, numeric-string, preference, voice-record, video-record, cart, qty-list - Improved UI for yes/no questions and code scanning - Generator File: Support set content-type and fields in downloadFile - Fix device ID issues when restoring from backup Breaking Changes: - Remove Generator TensorFlowInference - Event handling is now asynchronous and can be configured to run sequentially - Remove gpt3-tokenizer from JS Sandbox - Add locking mechanism in calculations (Disable it in Application Settings when editing) v2.21.10 - Local Debugging (Preview) - Fix connection issue in MCP Streamable HTTP transport - New MCP tools: update-brick & update-canvas - Generator LLM (GGML) - Bump llama.cpp runtime to b5849 - Support `Completion Enable Thinking` parameter for enhanced reasoning capabilities - Generator LLM (OpenAI-compat): Fix error handling - New Generator: Reranker (GGML) (Preview) - New Generator: VAD (Voice Activity Detection) (GGML) (Preview) - New Generator: Realtime Transcription (Preview) - Generator STT (GGML): Deprecate realtime transcription actions, please use Realtime Transcription instead - Generator Assistant (Preview): Improve message summarization for all LLM variants - Generator TTS (ONNX): Improve phonemizer support and add StyleTTS 2 (Kokoro) models - Generator File: Rename FILE_DELETE action for consistency - Generator MCP: Only reconnect if server supports /GET endpoint and bump MCP SDK - Brick Text: Fix rendering on line number 0 - System Properties: Add Device ID property for device identification - System Actions: Add DELAY action for timed operations

v2.21 New Features and Changes: - Add local debugging via Chrome DevTools Protocol. Open `Advanced Settings` to enable - Add new System Action: Data Expression - LLM context limit now defaults to 99 - New Brick: Rive (Preview) - Support for Rive animations - Brick Image: Improve performance and rendering with new image engine - Generator Question: Support more input types and improved UI - New types: language, numeric-string, preference, voice-record, video-record, cart, qty-list - Improved UI for yes/no questions and code scanning - Generator File: Support set content-type and fields in downloadFile - Fix device ID issues when restoring from backup Breaking Changes: - Remove Generator TensorFlowInference - Event handling is now asynchronous and can be configured to run sequentially - Remove gpt3-tokenizer from JS Sandbox - Add locking mechanism in calculations (Disable it in Application Settings when editing) v2.21.9 - Generator LLM (GGML) - Bump llama.cpp runtime to b5471 - Support audio input, currently only supported models from Ultravox - You can use `{ type: 'input_audio', input_audio: { data, format } }` message content for audio input - Fix cache K / V type properties not actually applied - Brick Video: Support actions - Brick Lottie: Support actions - Brick GenerativeMedia (Preview): Support default image hash and hash type - Brick Slideshow: Fix paths check of empty array for next state - Local Debugging (Preview): Add button for download logs on the DevTools home page - System Channel: Fix channel subscription not cleanup if subspace unmounted - Information: Improve layout for better display - Generator TTS (ONNX): Support OuteTTS - Support new Generator: TTS (GGML) (Preview) - Support OuteTTS v0.x models (v1.0 support coming soon)

v2.21 New Features and Changes: - Add local debugging via Chrome DevTools Protocol. Open `Advanced Settings` to enable - Add new System Action: Data Expression - LLM context limit now defaults to 99 - New Brick: Rive (Preview) - Support for Rive animations - Brick Image: Improve performance and rendering with new image engine - Generator Question: Support more input types and improved UI - New types: language, numeric-string, preference, voice-record, video-record, cart, qty-list - Improved UI for yes/no questions and code scanning - Generator File: Support set content-type and fields in downloadFile - Fix device ID issues when restoring from backup Breaking Changes: - Remove Generator TensorFlowInference - Event handling is now asynchronous and can be configured to run sequentially - Remove gpt3-tokenizer from JS Sandbox - Add locking mechanism in calculations (Disable it in Application Settings when editing) v2.21.8 - Local Debugging (Advanced Setting): Now moved from Experimental to Preview - Automation: Correct timezone for `cron` tests - Generator LLM (GGML) - Bump llama.cpp runtime to b5345 - Initialize support for multimodal (Preview), currently vision only - See supported models here: [(HuggingFace link)](https://huggingface.co/collections/ggml-org/gguf-vision-models-68244e01ff1f39e5bebeeedc) - Support new Actions: TOKENIZE and DETOKENIZE - Support new Outlets: TOKENIZE_RESULT and DETOKENIZE_RESULT - Generator LLM (OpenAI-compat): Ensure Claude / Gemini compatibility - Generator Assistant (Preview): Add IMAGE property to actions that create or update messages - Brick WebView: Support QUERY_SELECTOR action & outlet

v2.21 New Features and Changes: - Add local debugging via Chrome DevTools Protocol. Open `Advanced Settings` to enable - Add new System Action: Data Expression - LLM context limit now defaults to 99 - New Brick: Rive (Preview) - Support for Rive animations - Brick Image: Improve performance and rendering with new image engine - Generator Question: Support more input types and improved UI - New types: language, numeric-string, preference, voice-record, video-record, cart, qty-list - Improved UI for yes/no questions and code scanning - Generator File: Support set content-type and fields in downloadFile - Fix device ID issues when restoring from backup Breaking Changes: - Remove Generator TensorFlowInference - Event handling is now asynchronous and can be configured to run sequentially - Remove gpt3-tokenizer from JS Sandbox - Add locking mechanism in calculations (Disable it in Application Settings when editing) v2.21.7 - Generator MCP (Preview): Support connect via Streamable HTTP transport (/mcp) - Type now default to `streamable-http`, please note that if you want to use deprecated `sse` - Generator MCP Server (Preview) - Serve endpoint that support Streamable HTTP transport (/mcp) - Fix template params parse in direct-link - Support `Additional Params` for `detect-data-changes` type - Generator LLM (ONNX) (Preview): Support tool call - Generator Assistant (Preview): Support LLM generators: - LLM (ONNX) (Preview) - LLM (Qualcomm AI Engine) (Experimental)

v2.21 New Features and Changes: - Add local debugging via Chrome DevTools Protocol. Open `Advanced Settings` to enable - Add new System Action: Data Expression - LLM context limit now defaults to 99 - New Brick: Rive (Preview) - Support for Rive animations - Brick Image: Improve performance and rendering with new image engine - Generator Question: Support more input types and improved UI - New types: language, numeric-string, preference, voice-record, video-record, cart, qty-list - Improved UI for yes/no questions and code scanning - Generator File: Support set content-type and fields in downloadFile - Fix device ID issues when restoring from backup Breaking Changes: - Remove Generator TensorFlowInference - Event handling is now asynchronous and can be configured to run sequentially - Remove gpt3-tokenizer from JS Sandbox - Add locking mechanism in calculations (Disable it in Application Settings when editing) v2.21.6 - Generator Assistant (Preview) - Fixed warning log that unexpectedly displayed: `The LLM model does not support chat template (Jinja)` - Fixed LLM generator availability check - Fixed issue where tool_calls would proceed even when MCP was not enabled - Fixed `enabled` check (default: true) for linked MCP generators - Added support for new summary feature: - Implemented `Summary Messages` properties - Added `SUMMARY_MESSAGES` action - Generator LLM (GGML) - Fixed application crash that occurred during completion with tools - Fixed GPU availability check on unsupported devices - Generator Data Bank - Fixed application crash that occurred when no `properties` were provided