Introduction

TL;DR

• An open-source VRM avatar companion app
• Load a VRM model, connect any LLM provider, and bring your AI companion to life
• Voice, memory, and a relationship system built in
• Local-first — your data stays on your device
• Desktop app with transparent overlay mode for macOS, Windows, and Linux

What is Utsuwa?

Most AI companions today are either text-only chat interfaces or locked behind proprietary platforms. You don’t own the character and you don’t own the data.

Utsuwa takes a different approach. It gives you a 3D avatar that speaks, remembers your conversations, and develops a relationship with you over time. No accounts, no subscriptions. Your data stays on your device.

Connect any LLM provider you want — OpenAI, Anthropic, Google, DeepSeek, xAI, any OpenAI-compatible endpoint (OpenRouter, Together, vLLM), or a local model through Ollama or LM Studio. Add voice with ElevenLabs, OpenAI TTS, or a local TTS server. For voice input, use a local Whisper server, Groq, OpenAI, or your browser’s built-in speech recognition. Load any VRM model as your companion’s body. Everything is modular and swappable — and can run entirely on your own machine.

“Utsuwa” means “vessel” in Japanese — a container for AI to inhabit visually. The app is the vessel; you choose what goes inside.

Getting Started

Contributing

Utsuwa is open source and welcomes contributions. See the Contributing Guidelines for how to get involved.