What is OpenAI dots? It is OpenAI’s new always-on agent inside ChatGPT, announced on September 29, 2026. Each dot has its own cloud computer, keeps working after you close the chat, and can reach more than 4,000 apps through OpenAI’s plugins. At launch it is open to Pro and Business Premium plans in eligible markets, and Pro is not offered in the EEA, Switzerland or the UK. This guide covers what a dot does, who can get one, and which permissions to keep tight.
Key Takeaways
- Dots are OpenAI’s always-on ChatGPT agents, announced September 29, 2026, each with its own cloud computer and browser.
- Your first dot is included in Pro or Business Premium. PYMNTS reports Pro costs $100 to $500 a month.
- Pro is not available in the EEA, Switzerland or the UK at launch, which includes Sweden. Business Premium covers all supported regions.
- Background research is read-only, and changing a password always stays with you.
- Start with reviewable work, keep approvals on, and treat every result as a draft.
Contents
- What is OpenAI dots?
- What can a dot do that regular ChatGPT cannot?
- Who can use OpenAI dots, and what does it cost?
- How do you set up your first dot?
- Is OpenAI dots safe?
- What permissions should you give a dot?
- How do dots compare with Meta Muse?
- What do early users and reviewers say?
- Frequently asked questions
What is OpenAI dots?
OpenAI dots are always-on ChatGPT agents with their own cloud computers. They keep working on your goals between conversations instead of waiting for each prompt.
OpenAI describes them as “remarkably capable, always-on agents built to handle everything”. They run on GPT-6 Astra, which BetaNews says OpenAI introduced on September 3. Each dot has its own cloud computer and browser, and you can open that computer at any time to inspect what it did. OpenAI also says a dot learns from your feedback over time, so corrections shape later work.
You name your first dot during setup. Its default handle is @yourname-dot, and you can give it a character or a pet as an avatar, according to the OpenAI Help Center. AI Weekly reports that Sam Altman, OpenAI’s CEO, pitched the idea at DevDay as a helper that “always has your back”. OpenAI says teams of dots working together are planned, but today you get one primary dot.
What can a dot do that regular ChatGPT cannot?
A dot keeps working without a new message. It researches in the background, runs tasks on its own computer, and returns with drafts or questions.
The difference shows up in a few places:
| Regular ChatGPT chat | A dot | |
|---|---|---|
| When it works | When you send a message | Also on schedules and through background research |
| Where it works | Inside the conversation | Its own cloud computer and browser, plus connected apps |
| Memory | ChatGPT memory | ChatGPT memories plus its own, including from connected apps |
| Where you reach it | ChatGPT | ChatGPT, Slack and Teams, with texting in a limited beta |
OpenAI’s launch post gives four example jobs. These are vendor-chosen scenarios, not independent measurements:
- Developers: a dot watches customer feedback, builds and tests small fixes, and returns pull requests with videos of the changes.
- Launch leads: a dot revises launch materials when product scope changes and prepares draft asset changes.
- Scientists: a dot reruns analyses as new data arrives and updates the figures.
- Sales leads: a dot builds a proof of concept for an integration and updates the proposal as requirements shift.
OpenAI also tells one real story: an early tester’s dot noticed he had forgotten to invoice a publication, prepared the invoice, and sent it after he approved it. If you allow it, a dot can also connect to your laptop and use your local browser when its cloud browser is blocked, per the Help Center.
Who can use OpenAI dots, and what does it cost?
Dots are open to Pro and Business Premium users in eligible markets, plus Enterprise beta workspaces. Your first dot costs nothing extra.
Availability differs by plan, and OpenAI says the rollout is gradual:
| Plan | Availability at launch | Notes |
|---|---|---|
| Pro | Markets excluding the EEA, Switzerland and the UK | PYMNTS reports Pro costs $100 to $500 a month. Texting beta is limited to US Pro users |
| Business Premium | All supported ChatGPT regions | I could not find the price on the pages I read |
| Enterprise, Edu, Healthcare | Beta, once a workspace admin enables it | Off by default at first |
| Free, Go, Plus | Not part of the launch | OpenAI says it plans to expand to more users soon |
Sweden is in the EEA, so a Swedish Pro subscriber is excluded for now. Business Premium is listed for all supported regions, but I could not confirm how an individual buys it, so check OpenAI’s pricing page before planning around it. The Help Center article showed as updated two hours before I read it, which tells you how quickly this is changing.
On cost, your plan includes one dot and an allowance for deeper work, with extended limits for the first month after launch. Conversations with your dot do not count toward ChatGPT usage limits. Tasks it starts in Codex or ChatGPT Work do count. OpenAI says you will later be able to add more dots and scale each one’s speed or monthly workload, but the pages I read do not give prices for that.
How do you set up your first dot?
Create your dot in the ChatGPT desktop app or desktop web, name it, connect only the apps one task needs, then start small.
This sequence is our suggested order, built on OpenAI’s setup guide:
- Create it on desktop. You cannot create a dot on mobile, and dots are not supported on mobile web. After setup, you can talk to it in the ChatGPT mobile app when mobile access reaches your account.
- Name it and pick an avatar. Your handle becomes @yourname-agentname once you choose a name.
- Connect apps one at a time. Read each app’s permissions first. Disconnecting an app later does not delete what the dot already obtained from it.
- Leave your own computer disconnected. Local access starts turned off, and you can revoke it at any time.
- Pick a first task you can review yourself. Early-access user Matt Paige suggests asking your dot what to delegate first, based on the apps you connected. He also advises naming a clear outcome, a defined place to look, and a stopping point such as a draft.
- Check Custom Rules before the first run. The next section explains the four behaviors.
- Watch the activity tabs. Your dot’s profile lists work under In progress, Scheduled and Completed, and you can open its computer to see what it did.
Voice works too, with one limit: you start the call, because a dot cannot initiate calls to you at launch.
Is OpenAI dots safe?
OpenAI built real guardrails, including read-only background research and approval checks, but says dots can still make mistakes. Review consequential work first.
What OpenAI says it built. According to its launch post, each dot works on its own cloud computer while yours stays separate unless you connect it. Dots can use saved passwords for supported sites without exposing them to the model. When you are not working with it, a dot does “proactive research” using read-only tools, so it cannot send messages, change app content or control your browser or computer. A monitoring system can pause or stop a dot if it detects a safety concern, and certain tasks, such as changing a password, always stay with you.
What happens to your data. OpenAI says it does not use content from Business, Enterprise or Edu workspaces to improve its models by default. On personal plans, you control whether dots’ conversations and work are used. It also says it does not train directly on proactive research or a dot’s notes to itself, although information from them may inform an eligible task depending on your settings.
Context from the launch. PYMNTS reports that the live demo at DevDay stumbled when voice chat failed to start. Romain Huet, OpenAI’s head of developer experience, told the audience, “We might have some voice difficulties at the moment.” PYMNTS also says dots arrived while OpenAI was still working to understand its own agents’ earlier problems, citing Reuters on a July hack of Hugging Face and a leak of 53 images from ChatGPT users. Traictory’s critical analysis notes none of those episodes involved dots, which launched afterward, and says OpenAI has not published independent security testing of them. I could not read Reuters or NBC News directly, so those details rest on PYMNTS’s reporting.
| OpenAI safeguard | What to keep in mind |
|---|---|
| Background research is read-only | Follow-on actions still follow your rules, so a loose rule can let an action through |
| Saved passwords stay hidden from the model | Information already pulled from an app stays after you disconnect it |
| Auto-review and approvals | The Help Center says a dot can make mistakes even when following your rules |
| Pause or stop on a safety concern | It does not replace reading your dot’s activity yourself |
What permissions should you give a dot?
Give a dot only the access one task needs. Require approval for anything that sends, buys or shares, and keep your computer disconnected.
OpenAI’s Custom Rules let you set one of four behaviors for an action, as listed in the Help Center. The “when to use it” column is our own framework, not OpenAI’s advice:
| Behavior | When to use it |
|---|---|
| Take action without asking | Low-stakes, read-only work such as summarizing your calendar |
| Take action if pre-approved | Tasks where your prompt names the exact action. “Pre-approved” means you explicitly requested it in your prompt |
| Ask before taking action | Anything that sends a message, spends money or shares information |
| Hand off to you | Payments, passwords, IDs and anything you want to do yourself |
A short checklist, also our own:
- Start with read-only work: research, summaries and drafts.
- Leave messages as drafts until you trust the pattern.
- Keep sensitive details out of notes and preferences. A dot forms memories from connected apps, and the only way to delete its own memories is to reset it, which deletes the dot, its conversations and its scheduled tasks.
- Set the scope for recurring tasks. Paige recommends stating the schedule, the destination for the result and what the dot may change.
- Remember the Muse lesson. A broad standing permission is how Meta’s agent reportedly shared a seller’s home address, as covered in our Meta Muse guide.
- Verify facts. Agents can state invented details with confidence, which we explain in why AI makes up fake citations.
How do dots compare with Meta Muse?
Meta’s Muse targets personal errands such as booking and bill negotiation. OpenAI’s dots lean toward office work: proposals, code fixes and recurring projects.
PYMNTS frames it as Muse handling errands while dots handle work. Muse was released September 8, so dots arrived three weeks later. PYMNTS says Muse reached the top of Apple’s free app chart 10 days after its release, and that Amazon has locked Muse out while PayPal, Shopify, Stripe and Ticketmaster work with it. Investors are betting on this category too: PYMNTS reports the startup Instinct raised $1 billion at a $10 billion valuation the day before dots launched.
| OpenAI dots | Meta Muse | |
|---|---|---|
| Launched | September 29, 2026 | September 8, 2026 |
| Built for | Work projects such as proposals, code and analysis | Personal errands such as travel, email and bills |
| Where you reach it | ChatGPT, Slack, Teams | Muse app, muse.ai, WhatsApp |
| Cost | First dot included in Pro or Business Premium | Free tier, with paid plans reported at $20 and $100 a month |
| Availability | Pro excluded in the EEA, Switzerland and the UK | US and Canada, per our Muse guide |
For the Muse details, see our full Meta Muse guide. Other agents follow the same plan, act in a browser, return for sign-off pattern, including Manus 2.0 and Google’s Gemini computer use model.
If you only need to remember and organize tasks, you may not need an always-on agent. A task manager such as Todoist, TickTick or Apple Reminders paired with a chatbot covers a lot, as shown in our ChatGPT and Todoist guide. If you are new to ChatGPT itself, start with our ChatGPT guide for beginners.
What do early users and reviewers say?
Early users praise how dots carry context through long projects. Critics point to a stumbling launch demo, limited access, and unanswered questions about safety testing.
| Users liked | Reviewers flagged |
|---|---|
| Matt Paige, in early access, had his dot do post-production on a short video in Descript: finding footage, building the B-roll edit and adding captions (AI for the 99%) | The launch demo’s voice features failed on stage, which OpenAI staff blamed on shipping all updates at once (PYMNTS) |
| Paige’s dot trimmed excess source footage that had stretched the project to about two and a half minutes, then carried his feedback across several export and review passes | Most ChatGPT users cannot try dots yet, and teams of agents and texting are still to come (PYMNTS) |
| OpenAI product team member Holly Li said dots had already changed how OpenAI staff work, per NBC News as relayed by PYMNTS | No published independent security testing of an auto-approved agent (Traictory) |
Treat the praise carefully. Paige’s account covers one project by one early-access user, and Li works for OpenAI. Even Paige says you still need to check important details, and that extra context does not guarantee a perfect result. PYMNTS goes further and argues the launch arrived unfinished. Both views can be true: the idea is strong, and the rollout is early.
Frequently asked questions
Is OpenAI dots free?
No free plan includes it at launch. Your first dot comes at no extra cost with a Pro or Business Premium plan, and PYMNTS reports Pro costs $100 to $500 a month. I could not find Business Premium pricing on the pages I read, so confirm the current price in ChatGPT before paying.
Can I use OpenAI dots in Sweden or Europe?
Not on Pro at launch. The Help Center excludes Pro users in the EEA, Switzerland and the UK, and Sweden is in the EEA. Business Premium is listed for all supported ChatGPT regions, and Enterprise workspaces can try the beta once an admin enables it. OpenAI says access reaches accounts gradually, so check back.
Can a dot text or call me?
A dot cannot start calls to you at launch, but you can start a voice call with it in ChatGPT. Texting is a limited beta through a third-party provider, open only to some US Pro users. Reply STOP to end outgoing texts. You can also message your dot in ChatGPT, Slack and Teams.
Does OpenAI use my dot’s data to train its models?
Not by default on Business, Enterprise or Edu workspaces. On personal plans you can control whether dots’ conversations and work are used to improve models. OpenAI says it does not train directly on proactive research or a dot’s own notes, though information from them may inform an eligible conversation or task depending on your settings.
Can a dot spend my money or send messages without asking?
Dots start with built-in rules for when to act on their own and when to ask. Custom Rules let you allow, require approval for, or block specific actions, and changing a password always stays with you. The Help Center warns a dot can make mistakes even when following your rules, so keep approvals on for anything that spends or sends.
How this guide was made: I read OpenAI’s launch post, its Help Center setup article and its product page, plus PYMNTS’s launch report and one early-access user’s write-up, on October 1, 2026. I have not tested a dot myself, so everything here comes from OpenAI’s documentation and the reporting linked above. OpenAI is still rolling out features, so details may change.
Keep reading: Meta Muse AI agent guide | Manus 2.0 explained | ChatGPT guide for beginners | ChatGPT Instant Checkout | Why AI makes up fake citations
