Andrew Ng Launches OpenCoworker: Desktop AI Agent for Everyday Tasks
Key Takeaways
- ▸OpenCoworker is a cross-platform desktop AI agent supporting multiple LLM providers and local models
- ▸The application can perform real work: reading files, managing emails/Slack, and creating professional documents and reports
- ▸Users maintain data privacy with local execution and can use OpenAI, Anthropic, Google, or Ollama models
Summary
Andrew Ng has launched OpenCoworker, a desktop AI agent that performs complex tasks directly on users' machines. The application can conduct deep research, read and write files with permission, manage communications through Slack and email, and create deliverables like PDF reports and spreadsheets. It also supports scheduled automations such as daily news summaries.
OpenCoworker is built on aisuite, a lightweight Python library that provides a unified API across multiple LLM providers including OpenAI, Anthropic, Google, Ollama, and others. The platform is cross-platform, with downloads available for macOS 13+ (Apple Silicon) and Windows 10/11 (x64), and requires users to bring their own API keys or run models locally via Ollama.
The application emphasizes data privacy—all data remains on users' machines. The source code lives in the aisuite repository, serving as a reference implementation for building AI agent harnesses. aisuite itself comprises three layers: a Chat Completions API providing a unified interface across providers, an Agents API with tools and toolkits, and support for MCP (Model Context Protocol) servers.
- aisuite provides a unified API abstraction, allowing model swapping by changing a single string parameter
- The full source code is open source, enabling developers to build their own agent harnesses
Editorial Opinion
OpenCoworker represents a meaningful shift in how desktop AI assistants are built and deployed. By abstracting provider differences through aisuite's unified API, Ng has lowered the friction for developers to create cross-provider AI applications while giving end users control over both their data and their model choice. The emphasis on local execution and data privacy, combined with support for open models like Ollama, positions this as a serious alternative to proprietary cloud-based AI assistants. This could accelerate adoption of on-device AI agents in professional workflows.
