Utsuwa
Open-source AI companion app with a 3D VRM avatar, voice chat, and local long-term memory.
At a Glance
About Utsuwa
Utsuwa is an open-source AI companion from Juice Boxx Games LLC that gives a language model a 3D VRM avatar, a spoken voice, and persistent memory. It runs in the browser and as a beta desktop app for macOS, Windows, and Linux. Users bring their own API keys or run the whole stack offline with local models. The latest release listed is v0.21.4.
What It Is
Utsuwa is a companion platform rather than a plain chat window. A VRM avatar breathes, looks around, lip-syncs to synthesized speech, and shows chat replies as speech bubbles that track its head. Optional relationship mechanics inspired by Japanese dating sim games track mood and bond over eight stages, from Stranger to Soulmate.
Models, Voice, and Memory
Chat works with OpenAI, Anthropic, Google Gemini, xAI, DeepSeek, Ollama, LM Studio, and any OpenAI-compatible endpoint. Text-to-speech options include ElevenLabs, OpenAI TTS, Fish Audio, local OpenAI-compatible servers, and OmniVoice. Voice input uses a local Whisper server, Groq, OpenAI Whisper, or the browser Web Speech API. Memory is stored locally in IndexedDB. Facts are indexed with local embeddings (Transformers.js) so they are recalled by meaning, and a memory inspector and graph let users review them. Save files can be exported and imported.
Play Features and Desktop Mode
Users can load custom VRM models, upload VRMA animations, and show the companion photos for vision-capable models. Photo mode offers poses, backgrounds, filters, and stickers. AR mode places the avatar in a real room on WebXR devices such as Android Chrome. The desktop app adds a transparent always-on-top overlay and global hotkeys. MCP servers can be connected so the companion can call tools during chat.
Current Status and Licensing
The README describes the desktop app as beta and unsigned, so operating systems warn on first launch, and notes the project is in early development. Starting with 0.12.1 the project is licensed AGPL-3.0-or-later; earlier releases remain under MIT. Live2D support and hands-free voice mode are listed as planned.
Community Discussions
Be the first to start a conversation about Utsuwa
Share your experience with Utsuwa, ask questions, or help others learn from your insights.
Pricing
Open Source (AGPL-3.0)
Free, open-source AI companion with 3D VRM avatar, voice chat and local memory for macOS, Windows and Linux. Bring your own keys or run fully offline; BYOK provider costs are external.
- Licensed AGPL-3.0-or-later (OSI-approved)
- 3D VRM avatar with lip-synced voice
- Long-term semantic memory stored locally
- Bring your own model keys (OpenAI, Anthropic, Google, DeepSeek, xAI, Groq, Ollama, LM Studio)
- Fully offline option
Capabilities
Key Features
- 3D VRM avatar viewer with lip-sync
- Voice chat with TTS and STT
- Local semantic long-term memory
- Eight-stage relationship and mood system
- Fully offline option with local LLM, TTS, and STT
- Photo mode with poses, filters, and stickers
- AR mode on WebXR devices
- MCP tool calling
- Vision: show the companion photos
- Custom VRM and VRMA animation uploads
- Reminders and timers
- Desktop overlay mode (beta)
- Save data export/import
