Skip to content

Embodied VRM desktop companion

The head.
Not the brain.

YUI renders a living VRM character on your desktop. It senses text, voice, and screen context, then leaves all judgment to the Hermes backend you wire in.

YUI is the body.
Hermes is the brain.

YUI is the embodied frontend: rendering, sensing, and the I/O surfaces. That is its whole domain.

Judgment, persona, memory, and the agent loop belong to the Hermes backend you connect. YUI fires events and renders whatever comes back; it never decides whether or what to speak. That line is the system's load-bearing principle.

YUI · the head

Rendering + sensing + I/O

VRM character · lipsync · speech bubble · tool-status chip · text & voice input · screen capture · proactive trigger firing

Hermes · the brain

Judgment + persona + tools

Agent loop · memory · LLM reasoning · tool execution · emotion/motion cues · silence decisions, all delegated to the backend you supply

What you need to run YUI

One required piece, and it ships in the repo; everything else is optional. Start minimal; add a brain and a voice when you're ready.

Required

Core

  • VRM model file

    Your character. A default model is bundled; bring any VRM 0.x/1.0 file to replace it.

Optional

Add-ons

  • Chat backend

    The brain: any OpenAI-compatible endpoint, or a full agent like Hermes over the Responses API.

  • Expression broker

    Publishes the emotion/motion vocabulary a backend agent reads back to drive generate_express.

  • TTS provider

    Irodori TTS Server, or any OpenAI-compatible speech endpoint.

  • STT + VAD

    Whisper-based turn detection for voice input.

  • Reference voice

    A clip to clone the character's voice, uploaded to the TTS server from the panel.

  • Standalone Mods

    Independent MCP servers: memory, tools, integrations.

  • Custom expression set

    Map emotion cues to your VRM's own blend shapes.

Ready to wire it up?

The Getting Started guide walks through every required and optional piece.

Getting Started →