What Is ChatGPT Dots? OpenAI's Always-On AI Agent Explained
OpenAI's Dots is an always-on agent with its own cloud computer and browser. Here's how it works, what it can do, and where it's limited.
What is ChatGPT Dots?
Dots is OpenAI’s always-on AI agent, built into ChatGPT, that keeps working after you close the app. Instead of answering a single prompt and stopping, a Dot takes an ongoing goal, connects to the apps it needs, and keeps acting on your behalf, checking in only when it needs approval or judgment. It runs on OpenAI’s top-tier model, referred to in the reveal as GPT-6 Astra, and comes with its own computer and browser in the cloud.
TL;DR
- Dots is a persistent agent, not a chatbot feature. It continues researching, monitoring, and completing approved tasks after you close the conversation or turn off your device.
- It runs on its own cloud computer and browser, and can optionally get access to your local machine (off by default) through the ChatGPT desktop app.
- Dots connects to the apps and plugins you already use, including calendar, cloud storage, Slack, and Microsoft Teams, and uses that context to complete broader tasks.
- It has memory and proactive research capability, meaning it learns your preferences over time and can quietly gather information before you ask about it.
- Sensitive actions still require you: things like changing a password or moving money need you to step in directly, and some actions always require explicit approval.
- Access is limited at launch, rolling out to eligible Pro, Business, and some Enterprise plans, with the EEA, Switzerland, and UK excluded for personal Pro accounts initially.
- The launch lands right after rival “personal agent” products gained traction, including Meta’s Muse and startups like Instinct, Open Claw, and Hermes, putting OpenAI squarely into a new agent competition.
Plans first. Then code.
Remy writes the spec, manages the build, and ships the app.
How does Dots actually work?
A Dot starts with a goal you give it, not a one-off instruction. You tell it what you want handled, connect the relevant services (calendar, Slack, cloud drives, etc.), and decide which actions it’s allowed to take without asking first. From there, the agent works through the task over time, using its own cloud-based computer and browser to do things like update documents, check for new information, or prepare materials, and it comes back to you specifically when it hits a decision that needs your approval.
OpenAI gives users an “activity area” inside ChatGPT showing what the Dot is currently doing, what’s scheduled, and what’s already finished. That’s a structural shift from a chat log to something closer to a task dashboard. The design target isn’t a quick answer, it’s a responsibility that spans days: watching a calendar, following a project to completion, tracking research, or flagging updates from connected accounts that matter to you.
What can a Dot actually do for you?
Based on OpenAI’s own demonstration, Dots are pitched around ongoing work rather than discrete requests. Examples shown include updating a website using pre-approved designs, pulling the latest numbers into a board deck, mapping out the technical changes implied by a meeting’s notes, rearranging a calendar conflict, and even handling logistics like rebooking a vendor or signing kids up for after-school programs the moment registration opens.
The common thread is that a Dot is meant to sit on top of a domain (a project, an inbox, a calendar) and manage it, not just execute single commands. On the developer side, Dots also connect into Codex, meaning they can write and test code, flag fixes for review, and scan for vulnerabilities as part of the same ongoing-responsibility model.
How deep is its access to your apps and data?
Dots use whatever apps and plugins you’ve connected to your ChatGPT account, and OpenAI has described thousands of existing connectors across productivity and communication tools. Once connected, a Dot can read permitted information from a service like your calendar, Slack, or Microsoft Teams, and use that context to decide on next steps across a larger task.
Local computer access is a separate, optional layer. It’s off by default and has to be turned on through the ChatGPT desktop app. If you enable it, a Dot can work with local files, kick off coding tasks, and fall back to your actual browser if its cloud browser can’t finish something. That optional local access is the clearest signal of how far OpenAI wants this to reach: from a sandboxed cloud assistant to something touching your actual machine, if you let it.
Does a Dot remember things about you?
Remy doesn't build the plumbing. It inherits it.
Other agents wire up auth, databases, models, and integrations from scratch every time you ask them to build something.
Remy ships with all of it from MindStudio — so every cycle goes into the app you actually want.
Yes. OpenAI says a Dot retains context, learns from feedback, and builds an understanding of your preferences over time, including how you like work done and what counts as a good result. It also does what OpenAI calls proactive research: checking permitted, connected sources and saving private notes before you’ve even asked about a topic.
That research mode has a hard boundary. It cannot send messages, change information through a plugin, or control a computer on its own. It can find something relevant and then suggest a follow-up action that would use those more powerful abilities, but the decision to act stays with the user. This is the mechanism that’s supposed to make Dots feel less like you’re re-explaining yourself every session and more like working with someone who already knows the context.
What is a Dot not allowed to do without you?
OpenAI draws a clear line around the most sensitive actions. Changing a password or transferring money requires you to take over and complete the action yourself. Permanent deletion or installing software may require approval every single time, even if you’ve approved similar actions before. You can pre-approve certain repeated, low-risk actions (like sending a specific recurring message), but connecting an app does not hand a Dot blanket permission to do anything inside that account.
There are also practical limits outside the permission system. Websites can block automated agents outright, tasks can fail partway through, and the agent can simply get something wrong. OpenAI’s own guidance is that users should review consequential work rather than assume it’s correct, which is a tacit admission that reliability isn’t guaranteed yet.
Who can actually use Dots right now?
Access is restricted at launch. Dots are rolling out gradually to adults on eligible Pro plans, Business Premium, and selected Enterprise workspaces, with Enterprise access starting in beta and requiring an administrator to turn it on. Personal Pro access excludes the European Economic Area, Switzerland, and the United Kingdom initially.
Eligible users start with a single Dot, created through ChatGPT on desktop or the desktop website. For roughly the first month, usage won’t count against plan allowances, before new usage terms kick in. That’s a temporary arrangement, and it leaves pricing for ongoing, always-on agent usage genuinely unresolved. Separately, from OpenAI’s DevDay commentary, talking to a Dot is included in your plan, but the actual work it performs in ChatGPT or Codex threads draws on token usage like any other API call. In other words, the conversation is free, the labor isn’t.
Is Dots worth it compared to other agent products?
Dots arrives into a category that already has active players. Meta’s Muse built a following as a personal agent, and startups like Instinct (recently raising at a reported $10 billion valuation), Open Claw, and Hermes have been pushing the same “agent that manages your life” concept. OpenAI didn’t acquire Open Claw outright, but it hired the product’s creator, and that influence shows in how Dots is framed: less a tool you prompt, more a chief of staff you give standing responsibility to.
Whether that’s worth adopting now depends on factors OpenAI hasn’t fully answered yet: reliability of the agent’s judgment, the actual cost once the free usage window ends, how much sensitive account access you’re comfortable granting, and whether reviewing its work actually saves time compared to doing the task yourself. For now, limited availability and unresolved pricing mean most ChatGPT users are watching this category from the outside rather than using it directly.
Where is OpenAI taking this next?
OpenAI currently caps users at one Dot, but the stated direction is toward teams of Dots working together, potentially splitting research, communication, coding, and organization across separate agents under one person’s direction. The company is also pushing Dots outside the ChatGPT app itself, into Slack and Microsoft Teams, with more messaging integrations expected. In workplace channels, a Dot can post under its own identity, but only its owner can direct it, meaning a colleague mentioning your Dot can’t secretly assign it new work.
That expansion matters because it changes what “using ChatGPT” means. It’s less about writing prompts and more about assigning a standing responsibility, monitoring an activity feed, and correcting the agent when it misreads what you wanted. Whether that model sticks will come down to trust: how much people are willing to delegate, how often the agent gets it right, and how comfortable they are with a system that’s quietly learning how they work.
Frequently Asked Questions
What model powers Dots?
OpenAI’s reveal describes Dots as running on GPT-6 Astra, the company’s top-tier model at the time of launch.
Can a Dot access my computer?
Only if you turn that on. Local access is off by default and must be enabled through the ChatGPT desktop app; without it, the Dot works through its own cloud computer and browser.
Can a Dot spend money or change my passwords on its own?
No. OpenAI requires you to take direct control for the most sensitive actions, including password changes and money transfers, and some actions like permanent deletion may need approval every time.
Who can use Dots at launch?
It’s rolling out gradually to eligible Pro, Business Premium, and select Enterprise accounts, starting with one Dot per user. Personal Pro access does not currently include the EEA, Switzerland, or the UK.
Will using a Dot cost extra?
Talking to your Dot doesn’t consume token usage. The work it performs in ChatGPT or Codex threads does draw on usage, though OpenAI offered roughly a month of usage that doesn’t count against plan limits before new terms take effect.
