Open-LLM-VTuber lets you hold hands-free voice conversations with any large language model through an animated Live2D avatar, running locally on your own machine. It has more than 14,200 stars on GitHub and was actively updated in 2026.
Talk to the avatar and it listens, thinks and answers out loud โ and you can interrupt it mid-sentence, just like a real conversation, thanks to its voice-interruption support. The system is modular: plug in local models through Ollama or cloud APIs for the brain, choose from several speech-recognition and text-to-speech engines for the ears and voice, and the Live2D character provides expressions and presence on screen. It works across Windows, macOS and Linux, in web and desktop modes, and is popular both for AI VTuber streaming and as a personal desktop companion.
Much of the project’s charm comes from how configurable the character is. Persona and behavior are set through configuration files, Live2D models can be swapped, subtitles display what the character says, and input works by voice or text. Speech recognition options include engines such as Faster-Whisper and sherpa-onnx, while the voice can come from engines like Edge TTS or GPT-SoVITS for cloned anime-style voices, among many supported backends. A Docker deployment is also provided for running it on a home server, and an active community shares models, voices and setups for streaming.
Open-LLM-VTuber is free and open source. Running it with local models costs nothing beyond your hardware; cloud LLM or TTS backends are billed by their own providers.
VTubers and streamers, companion-app builders, and voice-interface experimenters who want a local, customizable AI character.