OpenAI Dots: What Day One Actually Includes

OpenAI launched dots on September 29, 2026 at DevDay: always-on agents in ChatGPT that run on GPT-6 Astra, get their own cloud computer and browser, connect to more than 4,000 apps through plugins, and keep working toward a goal after you close the chat. Your first dot comes with a ChatGPT Pro or Business Premium plan at no extra cost, and for the first month its usage does not count against your plan allowance. OpenAI's own example for creators: a dot that learns your voice, finds clip moments in a new interview transcript, and drafts show notes and social posts for your approval.

We rendered the six official pages behind the launch on launch night (the announcement, two Help Center articles, the safety post, the product page and the DevDay recap) and checked every claim a creator would rely on before connecting a dot to client work. Most of the launch holds up. Four details do not match between the announcement and the setup docs, no page says what a dot costs after the first month, and one privacy rule matters more than anything else on the list: you cannot delete a single thing your dot remembers. You can only delete the dot.

What OpenAI shipped on September 29

A dot is not a chat thread with a longer memory. It is a standing agent with its own Linux computer and Chrome browser that you can open and watch at any time, and it can run several projects at once. You give it a responsibility rather than a prompt, and it decides the next steps, messages you with progress or questions, and brings finished work back for review. Dots lead the DevDay recap, ahead of more than 20 other announcements across ChatGPT, Codex and the API.

Four mechanics define how it behaves:

• Proactive research. When you are not talking to it, a dot reads the apps you have connected with read-only tools and saves private notes. OpenAI says these research tasks are restricted in code: they cannot send messages, change app content or control a browser.
• Memory. A dot starts with your ChatGPT memory and builds its own, including from connected apps.
• Channels. You message or call it in ChatGPT on desktop, web and mobile, and can bring it into Slack.
• Local access, off by default. You can let it use your own computer through the ChatGPT desktop app. It can also start tasks in Codex cloud environments you have already created.

Independent coverage from SiliconANGLE and 9to5Google matches the announcement on model, plans and rollout. Access is gradual: the Help Center says it "may take several days to reach your account."

Who can get a dot, and what it costs

The ChatGPT pricing page lists "Dot, your always-on agent" under Pro and marks it "No" for Free, Go and Plus. Pro comes in three usage tiers at $100, $200 and $500 a month. On the business side, dots come with Business Premium seats, which cost $125 per user per month, or $100 billed annually.

Two rules decide the real cost. First, conversations with your dot do not count toward ChatGPT usage limits, but tasks it starts in Codex or ChatGPT Work do, "as usual." Second, the free allowance has a clock: "For the next month, dots usage won't count toward eligible Pro, Business, and Enterprise users' plan allowances. After this period, we'll share usage terms for each plan." That puts the end of the free period around October 29. OpenAI also says you will later be able to add more dots and pay for more speed or monthly capacity per dot. None of the six pages gives a price for any of it.

The regional split has a practical consequence. A creator in the UK or the EU cannot get a dot on Pro, but a Business Premium seat is sold in every supported ChatGPT region, so a one-person business workspace is the route in.

The announcement vs the Help Center

We ran the same pattern checks across all six pages and kept the rendered text and script with our measurements, so the check can be re-run when OpenAI edits the pages. Where the announcement and the setup documentation differ, the setup documentation is the one describing what your account can do today.

None of these is a reason to skip dots. They are the difference between planning around a Teams-based workflow that may not exist on your account yet and planning around Slack, which does. The Help Center also adds two limits the announcement never mentions: a dot cannot have its own standalone email address at launch (you can connect your personal email instead), and dots are not available to anyone under 18.

The memory catch: you cannot delete one thing

This is the rule to read before you connect anything. The dots privacy FAQ says it plainly: "You currently cannot view, delete or directly modify individual dot memories, including specific details that enter the dot's context from plugins."

Three more sentences in the same documentation close the other exits:

• Disconnecting an app "stops new access through that connection. It does not delete information your dot has already built into its context."
• Turning off ChatGPT Memory "stops that sharing. It does not delete information that your dot has already received."
• Proactive research can "review connected information proactively and form memories from it, even when you haven't asked a specific question about it."

The only control is Reset, which deletes the whole dot: its conversations, saved memories and scheduled tasks. Files, Codex threads and ChatGPT conversations the dot created are stored separately and survive the reset. OpenAI does say a dot's context keeps no credentials, images or screenshots.

For a creator this turns app connection into a one-way door. Connect a shared drive that holds one client's unreleased campaign, and the dot may read it in the background and keep what it learned until you wipe everything. The fix is order of operations, covered in the workflow below: decide what the dot should know before you connect, not after.

Training follows your plan. OpenAI does not train on Business, Enterprise or Edu workspace content by default. On personal plans, including Pro, the "Improve the model for everyone" setting decides whether a dot's conversations, actions and automations can be used. OpenAI does not train directly on proactive research notes, but a note that gets pulled into a conversation becomes eligible under that same setting.

How approvals and Custom Rules work

A dot ships with built-in rules for when to act and when to ask, and OpenAI's safety post describes three fixed tiers you cannot change:

• Always handed back to you: changing a password, transferring money between financial accounts.
• Confirmation every time: permanently deleting data, installing or running software from an unrecognized source, granting new security-sensitive access.
• Approval required: purchases with cards saved on a merchant's site. Approval can be given in advance if it covers that specific purchase.

Sharing depends on how sensitive the data is. Health data always needs a named recipient. Less sensitive details such as an email address need a class of recipient ("any airline company") by default.

Custom Rules sit on top. For a supported action you pick one of four behaviors: "Take action without asking," "Take action if pre-approved," "Ask before taking action," or "Hand off to you." Pre-approved means you asked for that action explicitly in your prompt. A dot can help you write rules but needs your approval to change them. Before anything that sends or changes content runs, a separate system called Auto-review checks the planned step against your instructions, your rules and OpenAI's safety requirements. Your rules cannot switch Auto-review off.

Sign-ins are handled with care. On supported sites, the model pauses while you type your password into a secure form that sends it straight to the browser, so the password never enters the model's context. That protection does not cover a password you paste into a chat or a document.

Dots vs Meta Muse

OpenAI is three weeks behind Meta here. Meta shipped its Muse agent on September 8 and, on the same day as DevDay, added Muse for Small Business with 15 named connectors including Canva, Figma and Shopify. Both products give each user an always-on agent on its own cloud machine. They differ on price, reach and who controls the approval switch.

If you already pay for Pro, a dot costs you nothing new for a month and plugs into the ChatGPT plugins, Codex environments and memory you have built up. If you do not, Muse is the cheaper place to learn what an always-on agent is good for.

How to set up a dot for a content workflow

OpenAI's creator example (clips, show notes, social drafts from a transcript) is a good first job because every output is a draft you review. Here is the order that avoids the memory trap, in about 30 minutes on the ChatGPT desktop app.

1. Pick the account. Client work belongs in a Business Premium workspace, where training is off by default. On a personal Pro plan, open Settings and turn off "Improve the model for everyone" before you create the dot.
2. Create the dot on desktop. The desktop app or desktop web is the only place you can create one. Name it; its handle becomes @yourname-agentname.
3. Write Custom Rules before connecting apps. Rules cover supported actions such as sharing, purchasing and access. Set sending email and posting anywhere to "Ask before taking action," and publishing to "Hand off to you." Rules apply to background work and delegated tasks too.
4. Connect the narrowest source you can. A dedicated folder of transcripts, not your whole drive. Whatever the dot can read, it may remember until you reset it.
5. Give it one responsibility with a finish line. For example: "When a new transcript lands in the Interviews folder, pick five clip moments with timestamps, draft show notes, and draft three social posts in my voice. Message me in Slack when the drafts are ready."
6. Leave local computer access off. It is off by default. The transcript job does not need it.
7. Watch Activity View for the first week. Check In progress, Scheduled and Completed on the dot's profile, correct it in the conversation, and note what it does unprompted.
8. Decide before the month ends. Usage is free until OpenAI publishes terms around October 29. Keep a list of what the dot saved you, so you can judge the price when it appears.

If something goes wrong, the Help Center says a dot may be able to reverse its own mistakes, such as undoing edits to a document or recalling an email, but "some actions cannot be undone." That is why sending stays on "Ask before taking action" until you trust it.

Frequently asked questions

What is an OpenAI dot?

A dot is an always-on agent in ChatGPT, launched September 29, 2026. It runs on GPT-6 Astra, has its own cloud computer and browser, connects to more than 4,000 apps through ChatGPT plugins, remembers context, and keeps working on goals between conversations.

Which ChatGPT plans include a dot?

Pro (all three tiers, $100 to $500 a month) and Business Premium seats ($125 a month, or $100 billed annually) include one dot. Enterprise, Edu and Healthcare get a beta that an admin must enable. Free, Go and Plus do not include dots.

Can I get a dot in the UK or the EU?

Not on Pro. The Help Center excludes the European Economic Area, Switzerland and the UK for Pro users. Business Premium users get dots in all supported ChatGPT regions, so a Business Premium seat is the route in.

How much does a dot cost after the first month?

OpenAI has not said. For the first month, dots usage does not count toward plan allowances, and OpenAI says it will share usage terms after that. Chatting with a dot never counts toward ChatGPT limits; Codex and ChatGPT Work tasks it starts do.

Can I delete something my dot remembers?

No. You cannot view, delete or edit individual dot memories. Disconnecting an app or turning off ChatGPT Memory stops new information but keeps what the dot already learned. Reset deletes the entire dot, including its conversations, memories and scheduled tasks.

Can a dot send email or post without asking me?

Only if your Custom Rules allow it. Each supported action can be set to act without asking, act if pre-approved, ask first, or hand off to you, and Auto-review checks sends before they run. Password changes and money transfers always go back to you.

Does a dot work in Microsoft Teams?

The announcement says so, but the Help Center setup guide names only Slack and texting as messaging channels. Plan around Slack until Teams appears in your dot's settings.