A routed AI runtime,
built from the cockpit up.
SYRIX began as a voice assistant and now runs as a local-first, provider-aware JARVIS/SYRIX system: brain routing, HUD widgets, voice, memory, live web, research tools, and guarded automation.
Phase-2B HUD contract
Widget commands travel through the SyrixBridge to hud_control.json, backend handlers update hud_state.json, and rolling acknowledgements keep UI actions visible.
13 specialist roles
NVIDIA NIM handles specialist roles when healthy, with Groq llama-3.3 cloud inference and Ollama local fallback routes available.
Approval-aware action
OpenClaw, guardian-shell previews, circuit breakers, and locked broker execution keep risky actions explicit and reviewable.
Real modules, not theatre.
The current codebase shows a practical assistant stack: local memory, voice/audio plumbing, Electron HUD state, live web providers, financial research, and safety gates.
Role-based routing
Main reasoning, coding, deep reasoning, vision, multimodal, rerank, and safety tasks route through configured model roles instead of one fixed model.
Electron cockpit
The HUD uses WebSocket state on port 5111 plus JSON bridge files for chat, control, diagnostics, widgets, memory, and system panels.
Voice and sound
PyAudio capture, Faster-Whisper STT, wake-word support, Shazam/music hooks, Piper chunked TTS, barge-in, and fallback voices.
Live provider stack
Open-Meteo, RSS, SearXNG, Wikipedia, DuckDuckGo, yfinance, Stooq, TradingView routes, maps, and optional Finnhub support.
Local-first memory
Short-term memory, JSON storage, Chroma fallback vector recall, markdown exports, and inspectable brain logs keep personal context grounded.
Research, not execution
Trading modules focus on watchlists, SMC/FI analysis, paper-mode reasoning, sentiment context, and circuit breakers. Broker execution is locked by default.
Avananka is built by Kabirr Kohli as a serious personal AI architecture: futuristic in interface, honest about providers, and guarded where real-world actions matter.
The K-7 rig is the target cockpit.
The project is tuned around a Windows desktop setup with RTX 4060 Ti, i7-12700K, 32 GB DDR5, microphone, speakers, and local/provider routes configured per module.
Access and support go direct.
For support, early access, or partnership questions, contact the project owner.