I built a voice-first AI assistant that lives behind a half mirror. Here’s how MirrorMate works, why it feels real, and how to host it yourself.

Noted's platform is  dedicated to providing insights and resources for note-taking, productivity, and knowledge management. Through articles, tutorials, and app reviews, Noted offers insights into effective note-taking techniques, digital tools for organizing information, and memory improvement strategies. Readers can learn about building a personal knowledge base, capturing ideas, and retaining information for better learning and productivity.

Noted

MirrorMate is a self-hosted voice-first AI assistant that runs behind a half mirror, combining hardware (Raspberry Pi, display, microphone) with software (Next.js, SQLite). It supports two deployment modes: minimal cloud setup using OpenAI APIs, or fully local using Ollama, VOICEVOX, and faster-whisper. The system features RAG-based memory for personal context, YAML-driven configuration for swapping LLM/TTS/STT providers, a plugin system for extensibility, and built-in integrations for weather, calendar, and reminders. The architecture keeps the Pi thin (UI and audio only) while offloading heavy AI workloads to a separate inference server, connected via Tailscale for privacy.

How I Built a Voice-First AI Mirror You Can Run at Home