OpenWorker: Desktop AI Coworker
openworker.com
OpenWorker is a new open-source AI agent that lives on your desktop and actually finishes tasks in Slack, email, calendar, and files. It runs on any model you pick, including local ones, and asks permission before doing anything risky.
Based on reporting by openworker.com — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
There's a specific kind of AI fatigue that comes from asking a chatbot for help and getting back a wall of text you still have to act on yourself. OpenWorker is built to skip that step entirely. Instead of answering questions, it takes on outcomes: prep me for a client call, untangle my Tuesday, tell me why checkout is throwing 500s. The tool then works through your actual accounts — Slack, email, calendar, HubSpot, files — and hands back something finished, whether that's a document, a rescheduled meeting, or a drafted incident report.
What separates this from the dozen other "agentic" pitches floating around is the deployment model. OpenWorker is fully open source, runs locally on your machine, and lets you plug in whatever model you want — OpenAI, Anthropic, Google, open-weight options like Kimi or DeepSeek, or a fully local model through Ollama. There's no OpenWorker subscription. You pay your model provider directly, and your API keys and OAuth tokens sit in a local secret store on your device rather than in someone else's cloud.
The examples the company shows off are the mundane stuff that actually eats a workday: a sales rep needing a renewal brief before a call, an on-call engineer trying to figure out which deploy broke checkout, a product team sorting through 47 feedback notes to find the theme that shows up 21 times. In each case the agent walks through multiple steps — reviewing logs, checking runbooks, drafting a message — and then stops. It surfaces a specific action, like emailing a CTO or posting to an incident channel, and waits for a yes.
That checkpoint is really the whole pitch. OpenWorker ships with four preset personas — sales, executive assistant, marketing, ops on-call — each scoped to particular tools and each configured to pause before anything leaves your machine. You can also trigger it passively: a Slack mention, an incoming email, a scheduled time, and it starts working in the background before you've even opened the app. Nobody's claiming this replaces a human coworker. But it's clearly aimed at replacing the fifteen small chores that make up the parts of a job nobody enjoys doing.
My take — AI-written commentary, not fact-checked reporting
The local-first, bring-your-own-model setup is the smart part here, not the agent framework itself — plenty of startups have tried to sell "AI that does your Tuesday for you" and stumbled on trust, not capability. Making the check-in step mandatory before anything leaves the device is the right default, and it's a little damning that so few competitors bother. If this catches on, expect the big AI labs to quietly copy the approval-gate pattern rather than admit an open-source side project got permission design right first.
Read more about this at: openworker.com
Related stories
Andrew Ng Just Released OpenWorker: An Open-Source, Local-First Desktop AI Coworker That Returns Finished Deliverables Instead of Chat
MarkTechPost · 1 month ago ·
40
Anthropic launches Cowork, a Claude Desktop agent that works in your files — no coding required
VentureBeat · 7 months ago ·
32
An opinionated guide to which AI to use to do stuff
Simon Willison's Weblog · 1 month ago ·
48