Andrew Ng Launches OpenWorker: An Open-Source Desktop AI Agent That Delivers Finished Work, Not Chat

Andrew Ng has announced OpenWorker, an open-source desktop AI agent that delivers finished work rather than conversation. The user specifies an outcome, such as a polished document or an updated calendar, and the agent breaks it into steps, works across local files and connected apps, and checks in before consequential actions. The architecture consists of four layers: a desktop shell (Tauri 2 with React 18), a local agent server (Python 3.10+ on FastAPI and uvicorn), a capability and connector layer, and a model router built on aisuite. The permission engine classifies tool calls into four risk classes (read, write_local, exec, external) and offers five permission modes: discuss, plan, interactive, auto, and custom auto. Unattended mode changes where the human is reached, not the autonomy ceiling. The privacy design is local-first: model calls go directly to the configured provider, conversations and tokens stay local, and secrets never enter the model's context. The only cloud component is an optional broker for OAuth handshakes. Model access is bring-your-own-key across 30 curated tool-calling models from providers like OpenAI, Anthropic, and Google, plus fully local models through Ollama. The repository contains 119 Python files and 149 TypeScript/TSX files.