second_brain

A personal AI system built around a persistent memory layer and a low-latency voice interface. It runs as two cooperating processes: a Python backend that owns memory and handles Telegram, and a Rust/Tauri desktop overlay that handles real-time voice conversation.

The system is intentionally personal — its personality, tone, and behavioral rules are defined in a plain-text soul.md file, not in code.

Vision

Most AI assistants start from scratch on every conversation. This one doesn't.

Every exchange is analyzed for durable facts, decisions, and open threads. Those are staged for review and — once confirmed — written into a structured memory graph of Markdown files. The next conversation begins with full context: who you are, what you're working on, what you decided last week.

The voice interface is designed for low friction. Push-to-talk, instant transcription, sentence-by-sentence TTS so the first word plays before the full response is generated. No typing required.