Familiar-Connect

An AI "familiar" that joins Discord voice channels, listens, understands speech, and talks back with real AI voices.

What's here

A Discord bot that reads and speaks in text and voice channels, backed by an event bus, layered context assembly, and a per-familiar memory store (Turso for relational data, tantivy for full-text search). Events flow through processors that assemble a prompt, call an LLM, and reply.

• CLI entry: familiar-connect run --familiar <id>.
• Discord text + voice subscriptions (/subscribe-text, /subscribe-voice, plus the unsubscribe-* counterparts).
• Twitch EventSub client.
• Turso history store (data/familiars/<id>/history.db) with facts, summaries, dossiers, reflections, and tantivy FTS + embedding retrieval.
• OpenRouter LLMClient, Deepgram STT, Azure / Cartesia / Gemini TTS.

Where to look next

• Running the bot: Installation → On-disk layout → Slash commands.
• Architecture: overview → memory strategies → voice pipeline.
• Activities: she gets up from the screen.
• Every knob, one page: Tuning.
• Prompting lessons: field findings.
• What's next: Roadmap.