What it does
OpenWorker is an open-source desktop app, currently in open beta, that runs an AI agent on your own machine. You describe an outcome, such as preparing a customer brief or triaging an inbox, and it breaks the work into steps, uses your files, terminal and connected apps, and hands back a deliverable such as a document, spreadsheet or Slack reply. It also supports scheduled automations like a morning brief or weekly report, and can be mentioned from a Slack channel. The app is bring-your-own-model. The page lists OpenAI, Anthropic, Google Gemini, DeepSeek, Kimi, Qwen, Mistral, Grok and others, plus open-weight models through Together and Fireworks and fully local models through Ollama. It connects to 25+ tools including GitHub, Slack, Jira, Notion, Gmail and Google Calendar, and to anything reachable over MCP. Control is a main theme. Writes, sends and shell commands are approval-gated, some dangerous operations are always left to the human, every tool call is logged with how it was approved, and commands can optionally run inside an NVIDIA OpenShell sandbox. Conversations, connector tokens and model keys are stored locally. Security-focused specialist agents, such as code vulnerability review, are the first ones shipped.
Capabilities
- Produces documents, spreadsheets, reports and Slack replies as finished deliverables
- Bring your own key for many hosted providers, or run local models with Ollama
- 25+ app connectors plus any tool reachable over MCP
- Approval gates, an audit trail and optional sandboxed command execution
- Scheduled automations with full transcripts
- Security specialists for code review, cloud posture and incident triage
When to use it
- Handing routine knowledge work such as briefs, reports and inbox triage to an agent that uses your own tools
- Reviewing a codebase and its dependencies for security risk, with proposed fixes held for your approval
- Teams that want to use their own model keys while keeping data and credentials on the local machine
