openworker: A Desktop AI Colleague That Delivers Results, Not Just Chat

This is a desktop AI agent focused on creating finished deliverables rather than just conversation. It delivers actual work results in the form of files or messages, ranging from code vulnerability scanning and security patch suggestions to document writing and schedule management. The core value lies in performing executable tasks beyond simple response generation.

It integrates with over 25 tools, including GitHub, Slack, Jira, and Gmail, to collect and process data. It supports OpenAI, Anthropic, and Google Gemini, as well as local models via Ollama, minimizing data leakage risks by allowing users to manage API keys directly. All integrations and model selections remain under user control.

Security governance is built into the architecture, requiring user approval for any arbitrary command execution or data modification. Dangerous operations are blocked even in auto-approval mode, and all actions are recorded in audit logs. The structure allows for gradually increasing autonomy by formalizing repetitive approval processes.

It particularly prioritizes security code support, combining static analysis tools with LLM inference to detect vulnerabilities and verify fixes. It is well-suited for developers and security personnel to delegate repetitive monitoring and initial response tasks. Currently in open beta, it is available for installation on macOS and Windows.