Andrew Ng Launches OpenWorker: An Open-Source AI Agent
Andrew Ng has launched OpenWorker, a free, open-source AI agent that does more than chat — it delivers finished work. Ng announced it on X on July 23, 2026, describing an agent that will “hand you a polished document, send a Slack message, or update a calendar entry” rather than just answer questions. He’s building it with Rohit Prasad, and the code is public at github.com/andrewyng/openworker under an MIT license. You can try it at openworker.com with your own API key.
The tagline is blunt about the pitch: “AI that gets your everyday tasks done.” Ask it to prepare a customer brief, untangle your calendar, draft a report, or triage a Slack alert, and OpenWorker works across your files and connected tools to produce the actual deliverable — then checks in before doing anything consequential.
What OpenWorker does
OpenWorker is positioned as an AI coworker, not a chatbot. The distinction Ng draws is the same one that separates a tool that helps you write an email from one that drafts it, files the follow-up, and updates your calendar. It integrates with everyday tools — Slack, email, calendar, Notion, and CRM systems — and the workflow is outcome-first: you describe the result you want, and it handles the underlying steps.
The design choices that make it notable:
- It produces deliverables. The emphasis is on finished output — a document, a sent message, a calendar update — not a suggestion you then execute yourself.
- It asks before consequential actions. OpenWorker “checks in before doing anything consequential, so you stay in charge.” The approval step is a core part of the design, not an afterthought.
- It’s local-first and privacy-preserving. It runs on your machine, and your data doesn’t leave it except through the LLM provider and integrations you explicitly choose. API keys and passwords stay in a local secret store.
- It’s model-independent. Bring your own key and run it on GPT-5.6 Sol, Claude Fable, Gemini 3.6, an open-weight model (Kimi, GLM, DeepSeek, Inkling), or Ollama to keep everything on your own hardware.
It runs on Mac today, with Windows support coming soon. The repository already has well over 500 stars within days of going public.
Why Ng built it
Ng and Prasad frame OpenWorker around a gap in the market rather than a new capability. AI coworkers, they argue, are becoming an important way to get work done — but the prominent ones are closed, cloud-hosted, and tied to a single provider’s model. Their stated goal is to give people an open, privacy-preserving, model-independent option: one you can inspect, run locally, and point at whichever model you trust or can afford.
That’s a pointed contrast with the direction the category has taken in 2026. Anthropic’s Cowork and OpenAI’s ChatGPT Work both run as hosted, cloud-based agents on the provider’s own models. OpenWorker’s answer is the opposite on every axis — local execution, open code, and your choice of model — with the trade-off that you bring (and pay for) your own API keys and run the software yourself.
Who it’s for
The honest read on OpenWorker is that it’s aimed at the technically comfortable and the privacy-conscious. “Bring your own API key,” “runs on your Mac,” and “keep your data local with Ollama” are features that land hardest with developers, security-minded teams, and people who specifically want to avoid handing their inbox and calendar to a third-party cloud. For that audience, a free, MIT-licensed, model-agnostic coworker is a genuinely appealing release — and the kind of open-source anchor that tends to push the whole category toward more openness.
For a non-technical user who just wants the work done without managing keys, model choices, or a local install, it’s a heavier lift than a hosted product. That’s not a knock — it’s the deliberate trade for control and privacy.
Where this fits
OpenWorker is the latest and most open entry in what’s become a crowded 2026 field of agents that do the work rather than describe it — alongside hosted options like Anthropic’s Cowork, OpenAI’s ChatGPT Work, and a range of assistants that act across your business tools. The shared thesis across all of them is the one Ng stated plainly: the value is in the finished deliverable, and in the agent knowing when to pause and ask.
Carly sits in the same “delivers finished work” category but takes the opposite delivery model from OpenWorker. Instead of software you run locally on your Mac, Carly is email-and-text native — you reach it the way you’d reach an assistant, from the Gmail or Outlook inbox you already use, with nothing to install and no API keys to manage. It connects to 200+ business tools and you delegate work in plain English. The two aim at different users: OpenWorker for the person who wants a local, open, bring-your-own-model coworker they fully control; Carly for the business owner who wants the work handled without running or maintaining anything. Carly’s pricing is free, unlimited Zapier-style workflows, with AI agents from $35/month.
FAQ
What is OpenWorker? OpenWorker is a free, open-source AI agent launched by Andrew Ng and Rohit Prasad on July 23, 2026. It works across your files and everyday tools — Slack, email, calendar, Notion, CRM — to produce finished deliverables like documents, messages, and calendar updates, checking in before it takes any consequential action.
Who made OpenWorker? Andrew Ng (founder of DeepLearning.AI and AI Fund) and Rohit Prasad. The source code is public on GitHub at github.com/andrewyng/openworker under an MIT license.
Is OpenWorker free? Yes. The software is 100% open source and free. You bring your own model API keys and pay the model provider directly — the OpenWorker project doesn’t charge you. Running local models via Ollama can avoid provider fees entirely.
What models does OpenWorker support? It’s model-independent. You can run it on GPT-5.6 Sol, Claude Fable, Gemini 3.6, open-weight models like Kimi, GLM, DeepSeek, and Inkling, or fully local models via Ollama.
What platforms does OpenWorker run on? It runs locally on Mac today, with Windows support announced as coming soon. It’s designed to be local-first, keeping your data on your machine except for the LLM providers and integrations you choose.
How is OpenWorker different from Cowork or ChatGPT Work? Anthropic’s Cowork and OpenAI’s ChatGPT Work are hosted, cloud-based agents tied to their makers’ own models. OpenWorker is open-source, runs locally, and lets you choose any model — trading the convenience of a managed cloud service for control, privacy, and model independence.
More: Best AI agents for productivity · Best AI tools for solopreneurs · Best AI personal assistants · What Carly can do
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
