Player not loading? Watch on YouTube
This tutorial builds a Jarvis-style voice assistant with LiveKit and Python, then adds camera input, web search and browser control. The presenter runs the agent and a custom frontend on a computer, but uses LiveKit Cloud and the Gemini API. Although he describes LiveKit as open source and says it can run locally, this walkthrough does not set up a self-hosted server or offline model inference.
Setup covers Python, uv, Node.js and the LiveKit CLI, with VS Code as the editor. The presenter edits the system prompt to give the assistant a sarcastic butler persona, then configures Gemini 3.1 Flash Live Preview and an API key. He describes his model access as free; that is an account-specific claim rather than a guarantee for every viewer. He also reports that the chat input does not work with this configuration.
A DuckDuckGo search function supplies web results, including weather information. OpenCode then uses the LiveKit documentation MCP server to help build Playwright browser tools. The workflow includes reviewing a plan before implementation and asking the coding assistant to fix a frontend startup error.
The final sections customize a local Next.js interface and deploy a Flutter mobile client. The presenter shows Flutter 3.47 and demonstrates Android setup with USB debugging. He discusses iPhone developer mode, but demonstrates the app on Android. The agent must remain running, and browser control is limited to the PC in this implementation.
