Purpose
Voice interaction, local model routing, data collection, and publishing had grown into separate programs with different controls and failure modes. Exocortex brings those parts into one observable operator console.
Project · active
Exocortex Local AI System
Local AI system that coordinates voice interaction, a room display, source collectors, publishing controls, and operator-visible state.
activeTauri + React + TypeScript schemasLocal Exocortex control plane with voice routing, source refresh, operator diagnostics, and scoped draft publishing.
My role
Designer and engineer
Status
active
Methods
- Tauri
- React
- TypeScript schemas
- Node sidecar
- Python Presence agent
- Piper TTS
- Windows SAPI
- Ollama / local LLM routing
- Room-display rendering
- Scoped Git publishing
Outputs
- The Tauri console owns settings, logs, source collectors, publishing controls, voice selection, and diagnostics. A local Python process handles wake phrases, microphone and speaker selection, Piper or Windows SAPI speech, deterministic command routing, local-model conversation, and a room display with listening, thinking, and speaking states. The Node sidecar collects approved activity sources and creates private field-note drafts for review.
Result
The system makes local agent behavior inspectable: operators can see routing decisions, source freshness, publishing state, voice configuration, and failures before allowing consequential actions.
Implementation
How it works
The Tauri console owns settings, logs, source collectors, publishing controls, voice selection, and diagnostics. A local Python process handles wake phrases, microphone and speaker selection, Piper or Windows SAPI speech, deterministic command routing, local-model conversation, and a room display with listening, thinking, and speaking states. The Node sidecar collects approved activity sources and creates private field-note drafts for review.
The working repository contains a Tauri and React console, a Node collector sidecar, shared TypeScript schemas, a Python room process, and a Next.js publishing target. Current work focuses on voice management, source diagnostics, scoped publishing, command routing, and reliable local operation.