Open-LLM-VTuber setup and local voice AI overview

Open-LLM-VTuber combines a voice assistant with an animated Live2D character. The speaker describes it as an open source, MIT-licensed project for Windows, macOS and Linux, with a browser interface and a desktop client that can display the character over other windows.

The walkthrough explains how microphone audio passes through voice activity detection, speech recognition, a language model and speech synthesis. Audio amplitude drives lip sync, while tags in the model's response trigger expressions. Configuration lets users choose engines independently, including Ollama or Llama.cpp for a local LLM and API providers for cloud inference. The speaker says an entirely local configuration can work offline; choosing cloud speech or model services changes that data path.

Setup covers cloning the repository, installing dependencies with uv sync, and copying conf.default.yaml to conf.yaml. That file selects the engines, avatar and character personality. The video also describes separate conversation state for each connection, saved chat history, screen or camera input, and MCP tools for an AI agent to call during conversation.

The memory discussion needs care: the speaker names persistent-memory integrations but later says long-term memory features are being reworked for version two. Version two is presented as an architecture rewrite in development, rather than a completed release.