Andrew Ng Launches Open-Source Desktop AI Agent OpenWorker

Andrew Ng, founder of DeepLearning.AI, is positioning OpenWorker as a desktop AI coworker that completes tasks rather than merely answering prompts. The open-source app can break an assignment into steps, work across local files and connected services, and return a document, spreadsheet or message as a finished deliverable. Its local-first, model-agnostic design is significant as businesses weigh the productivity gains of autonomous agents against concerns over data control, vendor lock-in and approval of consequential actions.

Ng unveiled OpenWorker on July 23, 2026, under the MIT license. The software uses a four-layer architecture: a Tauri 2 and React desktop shell, a local Python FastAPI agent server, a tools-and-connectors layer, and a model router. Built on Ng’s aisuite library, it offers a curated matrix of 30 models from 13 providers and can also connect to local models through Ollama. OpenWorker runs its agent loop, conversations and credentials on the user’s machine; users supply their own cloud API keys when choosing hosted models.