OpenWorker
#3 in Browser- & Desktop-AgentenAndrew Ng · 4× · tolest 25. Juli 2026
62
Momentum
OpenWorker is an Open Source desktop AI agent by Andrew Ng and Rohit Prasad that doesn't just discuss tasks but delivers finished work results—such as documents, Slack messages, or calendar entries. The application runs locally on the user's machine (Tauri desktop shell with local Python FastAPI server), can be used model-agnostic via bring-your-own-API-key, and additionally supports fully local models through Ollama. A typed risk and approval system secures critical actions such as file modifications, code execution, or external calls. The project was released as Open Beta under MIT license on July 23, 2026, and is available via GitHub and the project website openworker.com.
Momentum-Verloop
01.05.30.07.
Features
| Autonomy Level | Plans and executes tasks autonomously, but requests explicit approval before risky actions (write, exec, external); 4 risk classes, 5 permission modes |
| Channels | Desktop app (Mac/Windows) plus bidirectional Slack integration (mention @OpenWorker in a channel) |
| License | MIT license (open source) |
| Modalities | Text/documents as core modality; also supports voice input via a Rust-based speech recognition component |
| Platform | macOS (Apple Silicon, signed & notarized); Windows 10/11 available but unsigned; no Linux build |
| Price | Free & open source (MIT license); user only pays their own model API costs |
| Release Date | July 23, 2026 (open beta) |
| Tool/MCP Integrations | 25+ native connectors (Slack, GitHub, Jira, Notion, Linear, HubSpot, Outlook, Gmail, Google Calendar, etc.) plus any MCP server as a client |