Self-host mirrormate

MirrorMate places a conversational AI behind a two-way mirror driven by a Raspberry Pi. The wake phrase "Hey Mira" starts a hands-free dialogue. You can pick from several model backends — OpenAI or a local Ollama instance — several voice engines — OpenAI TTS or VOICEVOX — and several transcription options — the browser's built-in API, OpenAI's hosted Whisper, or a local Whisper build. A retrieval-augmented store remembers personal details from earlier chats and weaves them into new answers. An on-screen face rendered with Three.js cycles through eight expressive states and lip-syncs to the audio. Ready-made links pull weather from Open-Meteo, events from Google Calendar, search results from Tavily, and can post to Discord. A plug-in layer allows extra widgets or sensors such as a vision module that notices when you look at the mirror. The codebase uses Next.js 15, React 19, Node.js, SQLite via Drizzle, and Docker. A fully local deployment offloads the heavy models to a separate GPU box connected over Tailscale, while a cloud-backed mode needs only the Pi. The project is still early and has rough edges.