Andrew Ng, a pivotal figure in AI research, has co-launched OpenWorker AI, an open-source agent designed to execute tasks and deliver finished work directly from your desktop. Developed with Rohit Prasad, this new tool aims to provide an open, privacy-preserving, and model-independent option for AI-powered productivity.
Beyond Chat: Deliverable-Driven AI
Unlike conversational AI, OpenWorker AI focuses on tangible outputs. It can draft reports, prepare customer briefs, untangle schedules, or triage Slack alerts, interacting with your files and integrated applications. The agent breaks down complex requests into actionable steps, then checks in for user approval before executing any consequential actions like sending messages or modifying calendar entries.
The system runs locally on macOS, with Windows support anticipated soon. This local-first architecture ensures user data remains on the machine, only leaving through chosen LLM providers and integrations.
Flexible Model Integration and Privacy
OpenWorker AI is designed for model agnosticism, allowing users to bring their own API keys for various LLMs. Supported providers include OpenAI (GPT 5.6 Sol), Anthropic (Claude Fable), Google (Gemini 3.6), and open-weight models like Kimi, GLM, DeepSeek, and Inkling. For those prioritizing absolute data locality, it integrates with Ollama local LLM, a critical feature as open AI models become more readily available.
With over 25 integrations, OpenWorker AI connects to platforms such as GitHub, Slack, Jira, Notion, HubSpot, Outlook, Gmail, and Google Calendar. It can also manage recurring tasks through scheduled automations, providing full transcripts of all runs.
This initiative directly addresses the growing demand for open-source AI assistant solutions that prioritize user control and data privacy. The project is currently in open beta, with source code available on GitHub.
