-
Notifications
You must be signed in to change notification settings - Fork 0
OpenWorker
openworker is an open-source desktop application from Andrew Ng's organization that positions itself as an "AI coworker that lives on your desktop and delivers finished work, not just chat." Rather than answering questions in a chat window, it takes a task request, breaks it into steps, and carries those steps out across connected tools and local files, returning a completed deliverable.
The tool connects to more than 25 external platforms, including Slack, GitHub, Jira, Notion, and Gmail, and uses those connections to execute multi-step work such as preparing documents or managing a calendar. Consequential actions, like sending a message or running a command, require explicit user approval before they proceed, and the tool keeps an audit trail of what it did. It also supports scheduling so the same task can recur automatically.
The backend is built on aisuite, a Python library that provides a unified interface across large language model providers. This lets openworker run on OpenAI, Anthropic, Google Gemini, or local open-weight models through Ollama, rather than tying the tool to a single vendor. The desktop interface is a React and Tauri application, paired with a Python agent engine and connector layer, plus built-in speech-to-text input.
The repository was created on 2026-07-20 and had roughly 8,960 stars and 238 open issues as of this review, released under the MIT license. The high star count against a very short lifetime and a large open-issue backlog signals strong early interest paired with a project still being hardened. There is no first-person production use to draw on, which keeps this at Assess.