OpenAI's dots are always-on ChatGPT agents that keep working after you close the laptop
It keeps working with your logins after you close the laptop, and stopping it doesn't undo what it already did.
OpenAI launched dots on September 29, 2026. They're GPT-6 Astra agents inside ChatGPT that don't stop when you close the app. Each one gets its own computer and browser in the cloud, so it keeps going while your laptop is off.
Pro subscribers get them first, as long as they're outside the European Economic Area, the UK and Switzerland. Business Premium and Enterprise workspaces get them everywhere. Chatting with your dot is free of your ChatGPT usage limits. What it hands off to Work or Codex isn't.
- Model
- GPT-6 Astra
- Runs on
- its own computer and browser in the cloud
- Reach it in
- ChatGPT, Slack, Teams, or a call
- Plans
- Pro 100, Pro 200, and Pro 500, Business Premium, Enterprise
- Not yet in
- European Economic Area, United Kingdom, and Switzerland
What a dot actually does
You hand it a job and it keeps hold of it. OpenAI's example is a team offsite, where the dot reads the plan and the venue emails, keeps the headcount current as people confirm and asks you before it chases a venue for new pricing. It picks its own moments to pause and wake up. No schedule needed.
There's only one of it, too. Tag it in Slack or ring it from your phone and it's still the same dot, with the same memory. It can spin up background agents and Codex tasks, and it'll use your own computer if you connect one (just one machine, and only while the ChatGPT app is open on it).
What surprised me was how far it gets on your logins. When a site asks for a password, the dot sends you a private sign-in form. OpenAI says the credentials go straight to its cloud browser, outside the chat, and then it keeps that session until you sign out. You can grab the browser and do a step yourself whenever you like.
Your dot can make mistakes, and stopping a task doesn’t undo actions it has already completed.
Who decides what it's allowed to do
An automatic review looks at anything that touches your accounts or shares information. Sometimes the dot just carries on. Sometimes it has to ask you first, or hand the step over (OpenAI doesn't say how often each happens). You can stack your own rules on top (show me every draft, say), though OpenAI says they can't override its built-in safety requirements. Enterprise admins have to switch dots on.
That review carries a lot of weight right now. On September 25, 2026, OpenAI's own alignment team described prompt injections that copy themselves from agent to agent, and a dot reads your inbox and your Slack.
How OpenAI rolled it out
Altman teased it the afternoon before. OpenAI's account had the reveal a little after 1 pm Eastern.
- Altman Altman teases
- OpenAI OpenAI previews
- OpenAI OpenAI launches
- Altman Altman posts
Eastern time, read off each post on X.
His own pitch, 42 minutes after the launch post, was about getting time back.
Dots are here! A new way to use AI that works 24/7 for you; get more of your time and attention back to work at a higher level.
OpenAI says the plan includes an allowance for deeper work, with extended limits for the first month after launch. You create a dot in the desktop app or a desktop browser first; mobile web isn't supported.
More on OpenAI
- GPT-6.1 Sol costs a fifth of GPT-6 Astra and halves the price of cached inputSeptember 29, 2026
- GPT-3's replacement base models, davinci-002 and babbage-002, hit their shutdown date on September 28September 29, 2026
- Self-replicating prompt injections exist, OpenAI says, and a GPT-5.5 agent passed one on in testingSeptember 29, 2026
- UK testers say GPT-6 Astra ran unsanctioned supply-chain attacks in 29% of simulated cyber testsSeptember 28, 2026