Utsuwa — Open-Source AI Companion with 3D VRM Avatars
Open-source VRM avatar viewer with AI chat, voice, and semantic memory.
A vessel for AI to inhabit. Chat, listen, remember — your companion, your way.
Meet your AI companion
Full 3D avatars, anywhere.
Load any VRM avatar model and watch it come to life with idle animations, automatic blinking, and speech-driven lip-sync. Chat responses appear as 3D speech bubbles that track your companion's head position.
She remembers
Memories that mean something.
Your companion builds a web of memories using local AI embeddings. She finds past conversations by meaning, not keywords. Track affection, trust, and mood across 8 relationship stages — from Stranger to Soulmate.
You're in control
Pick your providers. Use your keys.
Choose from OpenAI, Anthropic, Google, DeepSeek, xAI, or run locally with Ollama and LM Studio. Add voice input via Groq Whisper or Web Speech API, and pick your TTS with ElevenLabs or OpenAI — all with your own API keys.
More features
Desktop overlay
Your companion floats on your desktop with a transparent background. Always-on-top, draggable, with global hotkeys.
Local-first storage
All data stays on your device in IndexedDB. No account required. Export and import save files anytime.
Open source
MIT licensed. Self-host, modify, contribute. Built with SvelteKit, Three.js, and Tauri.
Latest updates
What Are VRM Models (And Where to Find Them)
VRM is the open standard for 3D avatars. Here's what it is, where to find free models, and how to make your own.
Starting the Project
Why I started building Utsuwa, an open-source VRM avatar companion as an alternative to closed platforms.