Overview
OpenWorker is an open-source, local-first desktop AI coworker created by Andrew Ng (DeepLearning.AI, AI Fund) and released as a public beta in July 2026. Its defining idea is that AI should hand back finished work — a polished document, a filled spreadsheet, a drafted report, a Slack reply with the numbers, an updated calendar — rather than a chat transcript you still have to act on. You describe an outcome, OpenWorker breaks it into steps, works across your desktop, local files, and connected apps, and returns a shareable file. The architecture runs entirely on your machine: a Tauri 2 desktop shell, a Python FastAPI agent server, and aisuite (another Ng project) for a unified model layer, so no data leaves your device except to the model and integrations you choose. It is model-agnostic and bring-your-own-key, supporting OpenAI, Anthropic, Google Gemini, DeepSeek, Qwen, Mistral, xAI Grok, and local Ollama, with token cost billed by your provider. A typed permission engine classifies every action (read, write_local, exec, external) and gates writes, sends, and shell commands behind explicit human approval, with audit logging and rollback. Slack is bidirectional — mention @OpenWorker in a channel and the work runs on your desktop, returning as a thread reply.
Key Features
- Deliverables-first — returns finished documents, spreadsheets, reports, and web pages with full run transcripts
- Local-first privacy — agent loop, conversation history, connector tokens, and model keys stay in a local secret store; secrets never enter model context
- 25+ native connectors — GitHub, Slack, Jira, Notion, Linear, HubSpot, Outlook, Gmail, Google Calendar, plus terminal and local files
- MCP client — any Model Context Protocol server plugs in with per-tool control, extending the integration surface
- Approval-gated actions — writes, sends, and shell commands require check-in; unscheduled requests queue in an inbox
- Scheduled runs — morning briefings, weekly reports, and 24/7 channel monitoring kept inside the app with transcripts
Pricing
| Plan | Price | For |
|---|---|---|
| Open Source | $0 | MIT-licensed desktop app; bring your own model key or run Ollama locally |
| Local Ollama | $0 | Fully offline inference with no per-token cost |
| Cloud model usage | Pay per provider | OpenAI/Anthropic/Gemini/etc. billed by your chosen API |
Comparison
Versus Manus, OpenWorker keeps your code, emails, and customer data on your own machine instead of a remote VM, trading some autonomy for data sovereignty and auditability. Against OpenClaw, OpenWorker is a signed desktop product with 25+ built-in connectors and an approval UI, whereas OpenClaw leans on chat-app-native personal automation. If you prefer an assistant that lives inside your inbox and calendar rather than producing files, Lindy is the lighter alternative, but it lacks OpenWorker’s local-first, open-source guarantees.