Business AI
OpenWorker
Open-source local-first desktop AI coworker that returns finished work
OpenWorker is an open-source, local-first desktop AI agent from Andrew Ng. Instead of chat, it returns finished deliverables - documents, triaged inboxes, updated calendars - with an approval-gated permission engine and 25+ integrations. The agent loop and secrets stay on your machine.
GitHub stars
18,508
daily snapshots
2026-09-01 to 2026-10-13
What OpenWorker is
OpenWorker is an open-source desktop coworker that returns completed outputs rather than just chat. It is designed to get everyday tasks done using your everyday tools.
Key features
- Real deliverables - produces finished work, not just conversation.
- Slack integration - works from your existing team chat.
- Everyday tool access - connects to common apps and services.
- Scheduled runs - run tasks on a cadence.
- Ask-before-acting - confirmation for sensitive actions.
- Local-first - your data leaves your machine only through the model and integrations you choose.
- Model flexibility - supports many providers plus local models via Ollama.
Install
Download the macOS or Windows binary, or run from source with Python 3.10+, Node 20+, and Rust.
Use case
OpenWorker is a good fit for individuals and small teams who want an open-source agent for practical task completion. The local-first stance is a clear privacy win.
License
MIT.
Honest caveat
OpenWorker is newer and less battle-tested than business-automation incumbents. Evaluate it for personal or low-risk workflows before rolling it out to a team.
At a glance
Who's behind it
Andrew Ng
- GitHub stars
-
https://github.com/andrewyng/openworker
18,508 stars - updated 2026-10-13
- License
- MIT
- Platforms
- macos, windows
- Model support
- Bring your own model - 30+ models via OpenAI, Anthropic, Google, DeepSeek, Kimi, GLM, Qwen, plus Ollama for local
- Website
- https://openworker.com