AVA is a personal AI assistant with a VRM 3D model — she exists in a browser window, talks, moves, and helps. Part experiment, part companion, part tool.
Built because the off-the-shelf options were either too limited or too dependent on platforms that could disappear (or simply too expensive). AVA runs on infrastructure I control.
What she does
chat mode
Conversational AI with personality. Powered by Groq API for fast LLM responses, it's extremely cheap compared to other APIs or way more efficient than running it locally. I'd need 3x 5090 to make her as fast as current version has. Text-to-speech so she actually talks. Speech-to-text so she can hear me. She has Vision too so she can see her surroundings.
assistant mode
Extended context window for longer, more complex tasks. Less personality, more utility. Complete with Web Search and PDF reader for analysis. Same underlying model, different configuration.
VRM model
A 3D avatar using the VRM format — the same standard used in 3D VTubing, using BVH animation format. XR Animator is used to convert any MMD or FBX animations that I found online. Animations respond to conversation state.
Why VRM
VRM is an open format. Easily modified/customized with free tools like VRoid Studio. The model is portable, not locked to any platform. It can be swapped, modified, replaced. Custom contents can be easily found on BOOTH.
Notes
This is a personal project, not a product. No plans to release it publicly — it's built for one user and reflects that. For the lite version or demo, you can try it out by clicking here. The code is open source at github.com/trisanap/open-vrm-companion.