ActivePythonFaster WhisperPocket TTSOpenAI APIVRChat

Project Helion

A fully local chatbot that hangs out in VRChat. Run it on another account and it just exists there. Listens to people, talks back, and works with whatever LLM you want. Also pushes text to the chatbox over OSC. Everything runs on your own machine.

Coming soon

Currently private. Will be open-sourced once it's ready. Windows-first, uses uv for deps, and one config.yml controls the whole thing.

What it does

Helion is a chatbot you can drop into VRChat. Run it on a separate account, point it at an LLM, and it just hangs out in the world. Listens to people with speech recognition and talks back through a virtual cable. It also writes to the VRChat chatbox over OSC.

Works with LM Studio, Ollama, or anything else that speaks the OpenAI API. You can add memory so it remembers people, screenshots so it can see the game, or music controls if you want background tunes.


Capabilities

Modular

Every piece is swappable. ASR, TTS, LLM, memory, vision. Pick what you want, ignore what you don't.

Speech Recognition

Faster Whisper, Vosk, Parakeet, or Moonshine. Pick whatever fits your hardware.

Text-to-Speech

Pocket TTS and Qwen3 run locally. TikTok and Google Chirp are there if you want cloud voices. Interruptable mid-sentence.

VRChat

Sends audio through a virtual cable and writes to the chatbox over OSC. Wake word support so it only responds when called.

Memory

Remembers people and past conversations with SQLite or MongoDB. Pulls recent memories into the prompt automatically.

Extra Stuff

Vision via screenshots, web search, a music player with auto-ducking, and a web dashboard for monitoring.


Getting started

Clone the repo and run the installer. It handles uv, the venv, and all dependencies. First time you start it, a web configurator opens to walk you through setup.

# Clone and install
git clone https://github.com/HoppouAI/Project-Helion.git
cd Project-Helion
install.bat
# Run
start.bat