What ChatGPT dots actually do (and how to use yours without wrecking your workflow)

Picture an assistant who never clocks out, never loses context between conversations, and quietly keeps a browser open just to finish the task you mentioned three days ago. That's not a hypothetical anymore. It's a product, and it's called dots.

At DevDay 2026, OpenAI pulled back the curtain on a genuinely new category of AI product — not a chatbot, not a plugin, but a persistent agent that lives inside ChatGPT and keeps grinding on your goals whether you're watching or not. At DevDay 2026 in San Francisco, OpenAI officially introduced Dots, a new class of always-on autonomous AI agents built to execute workflows continuously without waiting for user prompts. If you've ever closed a ChatGPT tab mid-task and wished the work just kept happening, this is the feature that finally does that.

This guide breaks down what dots actually are, who can use them right now, and how to set one up so it does real work instead of just looking impressive in a demo.

What exactly is a ChatGPT dot?

Dots are always-on agents that live inside ChatGPT and run on GPT-6 Astra, OpenAI's flagship model. The distinction that matters most: a regular ChatGPT conversation waits for you to type something. Unlike a normal ChatGPT conversation that waits for the user to send another prompt, Dots are designed to continue working independently.

You can read the official breakdown on OpenAI's "Introducing dots" announcement page and the developer documentation at developers.openai.com/codex/dots, which walks through setup, messaging, tasks, memory, and computer/app connections.

Here's what separates a dot from the agent behavior you may already be using in ChatGPT:

Its own cloud computer and browser. The fundamental difference for ChatGPT's Dots is that they each have their own cloud computer and browser, giving them an environment within which they can interact with the broader web, use apps, run code and work with other plugins, keeping users' computers totally separate and free of any potential damage an autonomous agent could otherwise take.
Massive app reach. Each Dot is powered by GPT-6 Astra, gets its own cloud computer and browser, and can reach more than 4,000 apps through OpenAI's plugin ecosystem.
Multi-channel access. You reach them by messaging or calling a dot in ChatGPT on desktop, web and mobile, or messaging it in Slack and Teams. Texting is coming.
Memory that compounds. A dot remembers what you asked for last week. It keeps the work moving while you're off the clock. And it checks in when there's a decision that actually needs you.

It's also worth noting that a dot isn't just a renamed version of existing ChatGPT agent tasks. Earlier ChatGPT agent features ran a task when you asked and stopped. A dot is persistent: it keeps a goal, a computer, a memory, and connected apps, and it reports back on its own schedule. OpenAI says teams of dots working together is the longer-term plan; at launch you get one primary dot.

Who actually gets access right now?

This is the part that trips people up, because access is narrower than the marketing suggests. Dots are rolling out in ChatGPT to Pro and Business Premium subscribers in eligible markets. The Pro rollout excludes the European Economic Area, Switzerland and the U.K., and Business Premium users get dots in all supported ChatGPT regions. Enterprise, Edu and Healthcare workspaces can try a beta once an administrator turns it on, and it is off by default.

If you're on a Pro plan, there's good news on cost. A user's first dot is included in a Pro or Business Premium plan at no extra cost, with extended usage limits for deeper work during the first month after launch. Conversations with a dot do not count toward ChatGPT usage limits.

Just know that not every delegated task is free of usage limits forever. Tasks the dot hands off to Codex or ChatGPT's work features count against your normal plan limits, so it pays to keep an eye on what your dot is actually spinning up in the background.

If you're on regular ChatGPT Plus, don't expect it yet — rollout so far has prioritized the higher tiers, and broader availability is expected to follow in stages over the coming weeks.

How do you actually set one up?

Setup is refreshingly simple, at least to start. According to OpenAI's own documentation, the process looks like this:

Go to desktop, not mobile. Create your dot in the ChatGPT desktop app or a desktop browser. Create your dot on desktop first; mobile web is not supported. You can also jump straight to chatgpt.com/dots in a desktop browser to start the creation flow.
Let it introduce itself. Setup offers app connections and, in the desktop app, access to your computer. Your dot introduces itself and uses the context available to it to suggest ways it can help. You can start talking to it while it gets to know your work.
Name it and give it a face. Give your dot a name and choose its shape, color, eyes, glasses, and accessories. You can change these anytime. It sounds like a gimmick, but naming it matters functionally too — your dot starts with a handle such as @yourname-dot, and naming it updates the handle, for example, @tibo-alfred for a dot named Alfred. That handle is how you'll tag it in Slack or Teams later.
Connect the apps it actually needs. Don't connect everything on day one. Start with calendar, email, or whatever single tool the first task touches, and expand access as you build trust in what the dot does with it.
Give it a real first task. Skip the toy examples. Hand it something with a genuine deadline or recurring cadence so you can watch how it handles ambiguity.

Once it's running, you can keep talking to it across devices and tools. You can continue talking to your dot in ChatGPT on desktop or in the mobile app when the supporting update is available.

What can you realistically use a dot for?

OpenAI's own framing leans toward "handle everything," but the practical, grounded use cases are more specific. Suggested tasks include scheduling meetings, making travel arrangements, budgeting, debugging systems and writing code. The company also cited software developers using Dots to monitor customer feedback and scientists rerunning analysis on experimental data.

A few workflows worth trying first:

Recurring admin you keep forgetting. Think invoice follow-ups, expense categorization, or weekly status summaries. The anecdote OpenAI itself leads with is telling: an early tester forgot to invoice a publication. His AI agent noticed, prepared the invoice, and sent it once he approved. That's the sweet spot — bounded, recurring, and easy to verify before it goes out.

Content and research pipelines. If you publish regularly, a dot can sit between your raw material and your output. It can find clip moments in a new interview transcript and draft show notes and social posts for approval. Creators and marketers are a natural early audience here because the review step keeps a human in the loop without requiring you to do the grunt work.

Calendar and life-admin tracking. Because a dot keeps context across sessions, it's well-suited to anything you'd otherwise track in a messy spreadsheet or a dozen half-finished notes — a project tracker, a habit log, a planning document you update piecemeal throughout the week.

Developer monitoring loops. A dot can research in the background, run scheduled jobs like calendar reviews, message you in Slack or Microsoft Teams when it needs a decision, and hand coding work to Codex. That makes it a decent fit for watching dashboards, triaging issues, or flagging anomalies without you babysitting a terminal.

What are the safety guardrails?

Autonomous agents with their own browser naturally raise the "what could go wrong" question, and OpenAI built in several layers of restriction. Work done this way uses read-only tools that cannot send messages, change app content or control the browser or computer, and OpenAI's monitoring system can pause or stop a dot if it detects a safety concern.

On the model-safety side specifically, GPT-6 Astra was trained to refuse harmful requests, including biological and cybersecurity misuse, and human and automated red teaming targeted changing instructions, ambiguous requests, and attempts to push a dot past its permissions.

Practically, that means you decide the leash length. Keep sensitive actions (sending emails, making purchases, posting publicly) behind an approval step until the dot has proven itself on lower-stakes tasks. Check its activity log regularly during the first couple of weeks — not because it's likely to go rogue, but because catching a misunderstood instruction early saves you from untangling a bigger mess later.

How does this compare to other agent products?

Dots didn't arrive in a vacuum. After Grok Bot, Instinct, Muse this is another co-sign of this emerging category of persistent, always-on personal agents. The pitch across all of them is similar: stop treating AI as a search box and start treating it as a standing team member. What's different about OpenAI's version is the sheer scale of connectivity — through OpenAI's plugin system, a dot can reach more than 4,000 apps — plus the enterprise angle. OpenAI is working with Microsoft to integrate specialist dots with their enterprise governance and security controls in Agent 365, making it easier for enterprises to deploy dots inside their organizations.

That enterprise push matters if you're evaluating this for a team rather than personal use. It signals OpenAI is positioning dots as infrastructure, not a novelty feature that gets quietly sunset in a year.

Should you start with one dot or jump straight to a system?

Resist the urge to hand your dot ten tasks on day one. The better move is picking a single recurring pain point — your inbox triage, your weekly report, your project tracker — and running it for a week before expanding. Watch what it gets right without guidance, note where it asks unnecessary clarifying questions, and adjust your initial instructions accordingly.

Once that first task runs cleanly for a few cycles, layer on a second one. The compounding value of dots comes from memory and context accumulation — the more you work together, the more your dot learns your preferences, how you think, and what good looks like to you, and the magic of dots is when they bring you work done the way you would do it, sometimes before you even think to ask. That kind of trust doesn't build from a single flashy demo. It builds from repetition.

The agents are coming whether you're ready or not. The only real question left is which task you hand over first — and whether you're disciplined enough to actually review what comes back before it goes out the door.