This project provides a self-hosted, voice-first AI assistant designed to integrate into everyday objects, specifically mirrors, offering personalized interactions without requiring a separate screen or cloud dependency. It targets users interested in local AI, privacy-focused assistants, and novel physical interfaces, enabling quick access to information and conversational AI during routine daily checks.

How It Works

MirrorMate employs a voice-first architecture, running entirely locally using Ollama for LLM/VLM and embeddings, VOICEVOX for text-to-speech (TTS), and Whisper for speech-to-text (STT). It leverages Retrieval-Augmented Generation (RAG) for a personalized memory system, allowing the AI to recall and utilize information from past conversations. The system can be deployed minimally on a Raspberry Pi with an OpenAI API key or fully locally on a more powerful machine (e.g., Mac Studio with GPU) for complete offline operation, featuring a Next.js/React/Three.js frontend and Node.js/SQLite backend.

Quick Start & Requirements

• Primary Install:
  • Easiest (OpenAI API): docker run -p 3000:3000 -e OPENAI_API_KEY=sk-xxx -e LLM_PROVIDER=openai -e TTS_PROVIDER=openai ghcr.io/orangekame3/mirrormate:latest
  • Fully Local (Ollama + VOICEVOX): git clone https://github.com/orangekame3/mirrormate.git, cd mirrormate, docker compose up -d (after ollama pull qwen2.5:14b).
• Prerequisites: Docker, Ollama, VOICEVOX (for local setup). OpenAI API key (for easiest setup). A Raspberry Pi is suitable for minimal setups; a Mac Studio or other GPU machine is recommended for full local AI processing. Bun is used for development.
• Links: Documentation, Getting Started, and Releases are available via the project's GitHub repository.

Highlighted Details

• Voice-First Interaction: Activated via a wake word ("Hey Mira"), supporting multiple Speech-to-Text (STT) providers including Web Speech API, OpenAI Whisper, and local Whisper.
• Personalized Memory: Utilizes RAG to store and retrieve conversational context, enabling personalized responses and memory recall.
• Expressive Avatar: Features a lip-synced avatar with distinct animation states (Idle, Listening, Thinking, Speaking) to visually indicate the AI's status.
• Multi-Provider Support: Offers flexibility in choosing components: LLM (OpenAI, Ollama), TTS (OpenAI, VOICEVOX), STT (Web Speech API, OpenAI Whisper, Local Whisper), and Embedding (Ollama, PLaMo-Embedding-1B).
• Built-in Integrations: Includes support for Weather (Open-Meteo), Calendar (Google Calendar), Web search (Tavily), Reminders, and Discord sharing.
• Plugin System: An extensible architecture allows for the addition of custom widgets and sensor integrations, such as the Vision Companion plugin for eye contact detection.

Maintenance & Community

No specific details regarding maintainers, community channels (like Discord/Slack), or roadmap were provided in the README text.

Licensing & Compatibility

The project is licensed under the MIT license. This permissive license generally allows for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

The project is explicitly described as "Work in progress," with core features functional but noted to have "rough edges." Users should expect ongoing development and potential instability.