Commit History
-
feat: security hardening, transcription mode, 15 new features, Gradio UI
Khanna, Videh Rakesh Rakesh +1: Claude Sonnet 4.6Security (12 fixes): - SQL injection prevention in PRAGMA key (hex encoding + validation) - Placeholder API keys in .env.example - User ID spoofing protection with whitelist validation - WebSocket message length limit (10k chars) - CORS middleware with configurable origins - Auth query param deprecation warning - Rate limit memory leak fix + auth brute-force protection (5/min) - Tool argument validation and type coercion - Extended sensitive file blocklist (.pem, .key, .p12, id_rsa, etc.) - Generic error messages (no exception type leakage) - SQLite connection timeout - Auth endpoint rate limiting Transcription mode: - "Jarvis transcribe this" → continuous speech capture to Apple Notes - Timestamped chunks with punctuation restoration - Basic speaker diarization (Speaker 1/2) - Keepalive thread bypasses 10-min idle timeout - Overlay shows transcription state with chunk counter New features (15): - Echo cancellation — mute wake listener during speak() - Multilingual support — JARVIS_L…0191dfa
-
feat: cross-device support — all Apple devices, cloud tool delegation, PWA
Khanna, Videh Rakesh Rakesh- Fix uninstall permission errors (sudo for /Applications) - Fix mac app not opening (hardcode JARVIS_DIR in C launcher) - Fix mic staying on after 'bye' (standby mode, menu bar indicator) - Fix calendar permission spam (cached permission checks) - Fix app opening (alias mapping: camera→FaceTime, vscode→VS Code, etc.) - Add startup automation permission requests from JARVIS process - Server error retry prompt in conversation loop Cloud tool delegation: - Tag 78 macOS-only tools (osascript/subprocess) - Auto-delegate to connected Mac via WebSocket when server runs on Linux - Listener connects to server WebSocket, executes tools locally, returns results - 26 tools run directly in cloud (web search, weather, tasks, memory, etc.) iPhone/iPad/Apple Watch: - Siri Shortcut now hits HF Space (not localhost) — works on all Apple devices - Shortcut: dictate → POST /api/ask → speak response → notification - PWA: apple-touch-icons (152/167/180px), safe area insets, iPad layout - iOS install prom…2aa3ea6
-
docs: rewrite README with full project overview and install guide
Khanna, Videh Rakesh Rakesh81ba037 -
feat: JARVIS security, cross-device, intelligence & polish overhaul
Khanna, Videh Rakesh RakeshPhase 1 — Security: - Fix command injection: shell=True → shlex.split() + chaining blocklist - Preserve tool blocks in STM (no longer corrupted by text filtering) - Add user_id ownership validation on cross-device commands - Add token-based auth to web UI (Bearer / cookie / query param) - Fix wake word mic race condition Phase 2 — Cross-Device: - Cloud sync on every memory save via cloud_db - WebSocket push channel (/ws/device/{device_id}) for real-time delivery - Propagate real user_id through Telegram, WebSocket, REST - Command timeout/expiry for stale pending commands - Device heartbeat loop from listener (every 60s) Phase 3 — Intelligence: - Background task scheduler with job registry - Calendar watch, task reminders, routine alerts - Command pattern analytics (top tools, peak hours) - Complete LLM fallback chain across all backends - Rate limiting middleware (60 req/min per IP) Phase 4 — Polish: - SQLite encryption via JARVIS_DB_KEY env var - Rich monitoring dashboard at /dashb…e2a2dda
-
fix: improve Siri Shortcut removal in uninstall.sh
Khanna, Videh Rakesh Rakesh- Try CLI, AppleScript, and manual fallback to ensure shortcut is deleted018c7e2 -
feat: JARVIS.app WebView UI, full permissions, admin install
Khanna, Videh Rakesh Rakesh- jarvis_app.py: native macOS WebView app showing JARVIS web UI (no terminal) - install.sh: request admin privileges upfront, request ALL permissions (mic, camera, Bluetooth, accessibility, automation, contacts, location) - Info.plist: all TCC permission descriptions declared - JARVIS.app built and installed to /Applications during install - Opens System Settings for user to grant Full Disk Access - Auto-opens JARVIS.app after install - .gitignore: exclude JARVIS.app6433ae1 -
feat: unified listener with embedded menu bar, clap/snap detection, always-start listeners
Khanna, Videh Rakesh Rakesh- Merge menu bar (rumps) into jarvis_listener.py — single process for all triggers - Add clap/snap detection (double-clap or snap to wake JARVIS) - Remove separate com.jarvis.menubar plist — menu bar now runs inside listener - Always start wake listeners even if mic check returns silent (TCC grants to .app bundle) - Add LimitLoadToSessionType=Aqua to listener plist (GUI session only) - Clap detector auto-suppresses during active JARVIS interaction - install.sh updated: 2 services (server + listener) instead of 389f8ccf -
feat: OS auto-detect, onboarding spelling retry, device registration, cross-device context
Khanna, Videh Rakesh Rakesh- server.py: add _get_os_context() for OS-aware system prompt, device list in user context - jarvis_listener.py: rewrite onboarding with name spelling retry loop (2.5s silence), device cloud registration - install.sh: reset onboarding on fresh install, wait for server ready - llm.py: reorder backend fallback (OpenRouter first), strengthen tool call rules - tools/__init__.py: add 5 concrete tool call examples + CRITICAL RULES - tools/builtin.py: add AppleScript fallback for open_app - user_profile.py: include registered devices in user context - uninstall.sh: add __pycache__ cleanupf0ff151 -
chore: remove generated app bundle and shortcut from repo
Khanna, Videh Rakesh Rakesh6e98fdf -
fix: increase all timeouts to 180s, extract tools before STM, add multi-step tool chaining, differentiate listener errors
Khanna, Videh Rakesh Rakesh- All LLM backend HTTP timeouts: 120s → 180s (llm.py) - Listener/greeter/clap client timeouts: 60s → 180s - Extract tool calls BEFORE apply_stm() to prevent corruption - Log warning when tool-call regex misses potential matches - Multi-step tool chaining (up to 3 rounds) in REST and WebSocket - Differentiated error messages in jarvis_listener.py (timeout vs connection vs parse)f3ed77c
-
chore: snapshot current workspace state
Khanna, Videh Rakesh Rakesh4148d13
-
feat: add Hugging Face Inference API backend
Khanna, Videh Rakesh Rakesh- Add stream_huggingface() with auto-retry across fallback models - Support 6 free HF models (Llama 3.1, Mistral 7B, Qwen 2.5, Gemma 2, Phi-3, Zephyr) - Integrate into stream_response, query_model_full, backend detection - Add HF_API_TOKEN and HF_MODEL env vars to render.yaml and .env.example - Expose HF_FREE_MODELS in server.py /api/status endpoint - No new dependencies (reuses httpx)0ba3d55 -
chore: regenerate Siri shortcuts with working HF Space URL
Khanna, Videh Rakesh Rakeshd415f11 -
fix: catch exception-based errors in fallback path (not just yielded errors)
Khanna, Videh Rakesh Rakesh6382d74 -
fix: more robust error detection for backend fallback
Khanna, Videh Rakesh Rakesh5c9f256 -
fix: add fallback to free models when Claude credits exhausted in /api/ask
Khanna, Videh Rakesh Rakesh53e9cd2 -
fix: set PORT=7860 env var in Dockerfile for HF Spaces
Khanna, Videh Rakesh Rakesh5a5c4f8 -
feat: add device control, app automation, tests, and other updates
Khanna, Videh Rakesh Rakesh8576f52 -
fix: HF Space build - exclude macOS-only deps, fix port to 7860
Khanna, Videh Rakesh Rakesh- Create requirements-server.txt without mlx-whisper, sounddevice, webrtcvad, pyttsx3 - Update Dockerfile to use requirements-server.txt and expose 7860 - Fix app_port in README frontmatter to match - Remove duplicate websockets entry from requirements.txt227f1be -
fix: use valid HF color value in README frontmatter
Khanna, Videh Rakesh Rakesh9ec07cc -
merge: sync with HF Spaces remote
Khanna, Videh Rakesh Rakesha3e8db1 -
chore: add MONGO env vars to render.yaml
Khanna, Videh Rakesh Rakeshd0f7cc7 -
feat: add cloud DB, user profiles, daily routines & work session tracking
Khanna, Videh Rakesh Rakesh- cloud_db.py: MongoDB Atlas integration with SQLite fallback - user_profile.py: preferences, daily routines, work session management - tools/user_tools.py: 14 LLM-callable tools for user context - server.py: user context in system prompt, 6 new REST API routes - tests/test_user_profile.py: 14 unit tests (all passing) - requirements.txt: added motor, pymongo - README.md: documented new features and setupea9ee18 -
JARVIS v2 — 33 tools, battery efficient, PWA
Khanna, Videh Rakesh Rakeshbc497bf -
JARVIS v2 — full system control, battery efficient, multi-platform
Khanna, Videh Rakesh Rakesh +1: Claude Opus 4.6 (1M context)- 33 tools: volume, brightness, dark mode, WiFi, Bluetooth, media, apps, timer, screenshot, clipboard, calendar, reminders, files - Battery efficient: no continuous mic, uses hotkey + Siri Shortcuts - OpenRouter with auto-fallback across 7 free models - STM filtering + AutoTune for JARVIS-like responses - PWA support (install on phone as app) - Telegram bot integration - macOS native app (auto-start on boot, no Terminal visible) - Siri Shortcut trigger endpoint (port 8111) - Web UI with voice, streaming, tool execution Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>f7f801d -
Add Siri Shortcut API + Telegram bot
Khanna, Videh Rakesh Rakesh122c0ff
-
PWA: install as native app + auto wake word + keep alive
Khanna, Videh Rakesh Rakesh6ca6491 -
fix: colorTo metadata
Khanna, Videh Rakesh Rakeshb7e8e66 -
JARVIS v1 — autonomous AI assistant
Khanna, Videh Rakesh RakeshMulti-model racing, STM filtering, AutoTune, wake word detection, voice I/O, persistent memory, 10 built-in tools, responsive web UI.6e8f2e5 -
initial commit
Verified signatured1c7ada -
JARVIS v1 — autonomous AI assistant
Khanna, Videh Rakesh Rakesh +1: Claude Opus 4.6 (1M context)Multi-model racing, STM filtering, AutoTune, voice I/O, persistent memory, 10 built-in tools, responsive web UI. Supports Claude API, Ollama, and OpenRouter (free models). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>6c07b4f