Andrew Ng’s New AI Does Your Work For You

Forget the chat window. A new class of AI agent, OpenWorker, fundamentally redefines automation by shifting from conversational AI to outcome-oriented execution. Users request a specific deliverable, like 'draft a project brief from my recent emails,' and OpenWorker autonomously integrates with your tools and files to produce the finished work. It connects to platforms like Slack, Gmail, Notion, and Jira, executing tasks from start to finish.

This paradigm contrasts sharply with typical AI assistants. Those tools often provide information, code snippets, or raw data, leaving you with the extensive work of integration, execution, and formatting. OpenWorker, however, aims to deliver a polished document, a well-crafted Slack message, or an updated calendar entry directly, minimizing your post-AI effort.

Co-launched by AI pioneer Andrew Ng and Rohit Prasad in July 2026, OpenWorker embodies a deliberate product judgment. Ng’s involvement signals a strategic focus on solving the actual needs of knowledge workers who require completed tasks, not just suggestions. The open-source agent, which runs against local models and hit over 10,000 stars on GitHub, even earned the "repository of the day" distinction.

OpenWorker’s fundamental design principle is local-first architecture. The agent’s core logic, memory, and credentials reside entirely on your machine, not a remote server. This ensures that your conversations, agent execution, API keys, connector credentials, and local files remain private by default, never leaving your desktop unless you explicitly choose a cloud model or integration.

Crucially, OpenWorker integrates a sophisticated human-in-the-loop safety system, giving you ultimate control. Before executing any consequential action—like sending emails via Gmail, modifying local files, or running terminal commands—the agent checks for your explicit approval. Its typed risk engine uses four distinct risk types and five permission modes to govern operations, ensuring user oversight over sensitive system interactions.

This desktop-native, opt-in model directly addresses the primary security and privacy concerns that often deter enterprises and individuals from adopting cloud-centric AI agents for sensitive tasks. OpenWorker’s four-layer architecture, including a Tauri 2 + React desktop shell and local Python FastAPI agent server, keeps data on your machine. This fosters trust, enabling use cases previously deemed too risky, and supports fully local execution with Ollama or allows you to bring your own API keys for cloud LLMs, avoiding vendor lock-in.

OpenWorker champions ultimate model freedom, dismantling the walls of vendor lock-in. Users aren't tethered to a single AI provider; instead, you can plug in API keys for over a dozen cloud-based large language models, including powerhouses like OpenAI, Anthropic, and Google Gemini. For those prioritizing privacy or offline capabilities, OpenWorker offers a robust alternative: running 100% local models via Ollama, ensuring conversations and processing remain entirely on your machine. This model-agnostic design puts choice directly in your hands.

Beyond model flexibility, OpenWorker connects deeply with your existing digital ecosystem. It boasts an extensive library of over 25 integrations, linking seamlessly with essential workplace tools. Imagine an AI agent orchestrating tasks across:

• Slack
• Jira
• GitHub
• Notion
• Google Calendar

This comprehensive connectivity enables OpenWorker to execute complex, multi-step workflows, transforming fragmented tasks into cohesive, automated outcomes.

This powerful architecture thrives on transparency, thanks to its open-source foundation and MIT license. Developers gain the power to inspect every line of code, build custom connectors tailored to unique needs, and actively contribute to a community-driven platform. This collaborative approach fosters rapid innovation and ensures OpenWorker remains adaptable and extensible. For a deeper dive into its internals and to join the community, explore the andrewyng/openworker - GitHub repository, which quickly amassed over 10,000 stars and earned Repository of the Day.

OpenWorker isn't a future promise; it's a present reality. A fully usable open beta is available for macOS (Apple Silicon, macOS 12+), with a Windows 10/11 (x64) build actively in progress. Its rapid traction is undeniable, quickly amassing over 11.7k stars on GitHub and earning "Repository of the day" status for its innovative approach to AI agents. This momentum underscores a broad interest in desktop-first, privacy-preserving AI.

Expert testing validates OpenWorker's practical utility and robust design. Analysts consistently praise its four-layer architecture and model-agnostic flexibility. Evaluations confirm the agent effectively handles approximately 80% of common knowledge worker tasks, from drafting project briefs from emails to updating calendar entries or triaging customer support tickets. This demonstrates a significant, tangible leap in personal AI productivity.

This new class of AI tools thrives on community engagement. Experience OpenWorker's transformative capabilities firsthand: visit the official website to download the app for macOS. Dive deeper into its MIT-licensed code on the GitHub repository, contribute to its development, and help define the future of outcome-oriented automation. Become part of a movement building truly autonomous, user-controlled AI.

What is OpenWorker?

OpenWorker is an open-source, local-first AI agent co-launched by Andrew Ng. Instead of just chatting, it connects to your tools (like Slack, Gmail, and Notion) to autonomously complete tasks and deliver finished outcomes, such as polished documents or updated calendars.

Is OpenWorker free?

Yes, OpenWorker is free and open-source under the MIT license. It's designed for you to bring your own API keys for cloud LLMs or connect to local models via Ollama, giving you complete control over costs.

What makes OpenWorker different from other AI agents?

Its key differentiators are its local-first architecture for privacy, its model-agnostic approach that prevents vendor lock-in, and its robust human-in-the-loop permission system that requires user approval for important actions.

What platforms does OpenWorker support?

OpenWorker is currently available as a signed and notarized application for macOS (Apple Silicon, 12+). A build for Windows 10/11 is also available, with official code-signing in progress.