OpenWorker

Quick Take

OpenWorker is an open-source, local-first desktop AI coworker created by Andrew Ng, founder of DeepLearning.AI and AI Fund. Its core idea is returning finished deliverables rather than a chat transcript: a written document, a Slack reply with the actual numbers pulled and checked, an updated calendar, or a triaged inbox, rather than a suggestion you still have to act on yourself.

What Makes It a “Local” Open Alternative

Most AI coworker products from major labs are closed-source, cloud-hosted, and locked to that provider’s own models. OpenWorker takes a different approach on all three counts:

• Open source (MIT license): the code is public, inspectable, and modifiable, hosted on GitHub with active development.
• Local-first: agent loops, conversation history, tokens, and API keys stay in a local secret store on your machine. Only the OAuth handshake for connecting a third-party integration routes through a cloud broker.
• Model-agnostic: bring your own API key for OpenAI, Anthropic, Google, DeepSeek, or other providers, or run entirely local models through Ollama with no per-token API cost at all.

Architecture

OpenWorker pairs a Tauri-based desktop shell (React UI, Rust wrapper) with a local Python agent server built on the aisuite library for unified access across model providers, plus a Rust-based speech-to-text sidecar for voice input.

What It Integrates With

OpenWorker lists more than 25 integrations spanning communication (Slack, Gmail, Outlook), productivity and project tools (Notion, Linear, Monday.com, GitHub, Jira, Asana), calendars (Google Calendar, Outlook), CRM (HubSpot, Attio), and file storage (Dropbox, Box, Google Drive), plus terminal access and local file management. It’s also extensible through MCP for connecting additional tools beyond the built-in list.

How It Compares Conceptually

OpenWorker sits in the same general category as cloud-based AI coworker and personal-assistant features from major labs, and as broader personal AI assistant tools, but its specific positioning is openness and local control rather than matching any particular closed product’s exact feature set. If your priority is self-hosting, model choice, and keeping your data and credentials off a vendor’s servers, that’s the gap OpenWorker is built to fill. If you want the most polished, zero-setup experience with the deepest integration into one company’s ecosystem, a closed cloud offering may still be the simpler choice.

Honest Limitations

• Open beta, actively evolving. Expect rougher edges and more frequent changes than a mature commercial product.
• You manage your own model costs and keys. OpenWorker doesn’t include model access, you supply API keys or local compute yourself, which is more setup than a bundled cloud product.
• Self-hosting means self-support in practice. Community-driven support isn’t the same as a vendor’s dedicated support channel.
• Permissions matter. With 25-plus integrations able to touch real accounts, email, calendars, files, review what access you’re granting and to which accounts before connecting anything.
• Installation requires developer comfort for anything beyond the packaged app. Building from source needs Python, Node, and Rust toolchains.

Alternatives Worth Knowing

• Ollama, for running fully local models that OpenWorker can use as its backend
• LM Studio, another local model runner with a GUI
• Cloud-based AI coworker and personal-assistant features from major labs, for a more polished, zero-setup alternative if openness and self-hosting aren’t priorities

For more on running AI locally and keeping data on-device, see Local AI Explained and Portable Context.