OpenAI Dots are AI agents that keep working in ChatGPT after you close the chat. Announced at DevDay 2026 on September 29, a dot is an agent that stays on. It watches the apps you connect, picks up recurring work, and comes back to you only when a decision needs you.
That is a big change for anyone who runs a business without a back office. It also raises fair questions. What can a dot actually touch? What does it cost? What happens to the data it reads?
Every answer below traces back to OpenAI's launch post and help center, followed by a setup walkthrough and three briefs you can hand a dot today.
What are OpenAI Dots?
OpenAI Dots are always-on AI agents that live inside ChatGPT. Each dot runs on its own cloud computer, uses the apps you connect, remembers context between conversations, and keeps working toward goals you give it. It brings results back for review and asks you before decisions that need your judgment.
OpenAI calls them dots with a lowercase "d", and each user starts with one primary dot. You can name it and pick a character for it. Its default handle is @yourname-dot, according to OpenAI's getting started guide.
Dots run on GPT-6 Astra, OpenAI's newest frontier model. Through OpenAI's plugin ecosystem, they can connect to more than 4,000 apps, though the apps available to you may vary.
The idea is delegation, not conversation. You do not prompt a dot for every step. You hand it a responsibility, and it decides what needs to happen next.
How OpenAI Dots work
A dot combines four things a regular chat does not have: its own computer, access to your apps, persistent memory, and permission to work when you are away. Each one changes what you can hand off.
1. A dedicated cloud computer and browser
Every dot gets its own cloud computer with a browser. It can work on several projects at once on that machine, and you can open the computer at any time to see what it is doing.
Your own laptop stays separate by default. To change that, connect it through the ChatGPT desktop app and confirm Allow access. The dot can then work with your local files and use your local browser when its cloud browser is blocked. Revoke access cuts it off again.
2. Connected apps through plugins
A dot uses the same plugin connections as ChatGPT and Codex, managed in ChatGPT's Plugins tab. If you have already connected your email or file storage to ChatGPT, your dot can use that connection within the permissions you granted.
You can also message your dot from Slack or Microsoft Teams. Context carries across channels, so a task you start in ChatGPT can continue in Slack without repeating yourself.
3. Memory and learning from feedback
Your dot receives memories from ChatGPT and builds its own, including from connected apps. The more you correct it, the better it learns your preferences and "what good looks like," according to OpenAI.
That memory is a one-way door for now. You cannot view or edit individual memories in your dot's own context, and the only way to clear them is to reset or delete the dot. Shared ChatGPT memories are separate and stay manageable in ChatGPT's Memory settings.
4. Proactive research when idle
When nobody has given it a task, a dot looks for ways to help, a mode called proactive research.
The tools it uses for this are read-only. During proactive research, a dot cannot send messages, change content in your apps, or control a browser or computer. Any follow-up action goes through the normal approval rules.
OpenAI Dots vs regular ChatGPT
Regular ChatGPT answers when you ask. A dot keeps going when you stop. That gap is the same one that separates a chatbot from an agent, which we cover in more depth in AI agent vs chatbot.
Regular ChatGPT vs a dot, based on OpenAI's launch post and help center
One catch on usage: a dot can start tasks in Codex or ChatGPT Work for you. Those tasks still count toward your plan's usage limits, per OpenAI's launch post.
How to set up ChatGPT Dots
You set up a dot on a computer, not your phone. Confirm you are on Pro or Business Premium in an eligible region, open the ChatGPT desktop app or ChatGPT in a desktop browser, and follow the onboarding prompts. Then connect apps, review the default rules, and give it its first responsibility. After setup, you can message it from the ChatGPT mobile app.
- Check your plan and region. Dots are rolling out to Pro users outside the EEA, Switzerland, and the UK, and to Business Premium users in all supported ChatGPT regions. Enterprise, Edu, and Healthcare workspaces get a beta that an admin has to turn on. Rollout is gradual, so access can take several days to reach your account.
- Create your dot on desktop. Use the ChatGPT desktop app on macOS or Windows, or ChatGPT on desktop web. Mobile apps and mobile browsers cannot create a dot.
- Name it. Pick a name and a character or pet. Naming it changes its handle from @yourname-dot to @yourname-agentname.
- Connect your apps. Open ChatGPT's Plugins tab, or Customize, then Plugins in your dot's profile on mobile, and connect the apps the dot needs. Review each app's permissions before you connect it.
- Review the default rules. Open Custom rules and decide which actions the dot can take alone. More on this in the permissions section below.
- Connect optional channels. Add Slack, connect your own computer, or set up texting if your account has the beta. All of these are configured from desktop.
- Give it a first responsibility. Start small, with one recurring job you can easily check.
How to give your dot a responsibility, not a one-off task
A dot earns its keep on work that repeats. "Draft a reply to this email" is a chat request. "Keep my client follow-ups from slipping" is a responsibility, and that is what a dot is built for.
A useful brief covers five things:
- The goal: the outcome you care about, stated once.
- Where to look: the specific apps, folders, or labels in scope.
- The standard: what a good result looks like, in your words.
- Its authority: what it may do alone and what it must bring to you.
- The check-in: when and how it reports back.
Be precise about authority. OpenAI's privacy and safety FAQ says approving one message does not give a dot ongoing permission to contact people for you. If you want recurring messages sent, say who receives them, what they say, and when.
Here are three briefs written for small business owners. Swap in your own tools.
Unbilled work check
Every Friday at 4 pm, compare the projects marked delivered in my project tracker against invoices in my accounting app. For anything delivered but not invoiced, draft the invoice using the client's last invoice as a template. Do not send anything. Message me a list with client names and amounts.
New lead follow-up
Watch the Leads label in my email. When a new inquiry arrives, draft a reply using our pricing doc and the client FAQ in Drive. Ask me before sending. If a lead has had no reply from me in 48 hours, flag it in Slack.
Weekly stock check
Every Monday morning, compare the stock sheet against the reorder levels in column F. Draft purchase orders for anything below its level and hand them to me for approval. Never place an order yourself.
These mirror an example OpenAI shared at launch, where an early tester's dot noticed a forgotten invoice, prepared it, and sent it only after approval.
What a dot can do for a small business
A dot is strongest on recurring work that spans several apps and is easy for you to check. OpenAI's own launch examples lean toward developers, product leads, and enterprise sales teams. The same pattern fits a five-person agency or a solo founder.
Where an OpenAI dot fits in small business work
The right column matters more than it looks. A dot is an operator. It can only work through the apps you connect, so its value depends on whether the systems behind your business already exist.
For a broader comparison of agent options built for owners, see our roundup of AI agents for small business.
Permissions, approvals, and safety controls
You control a dot at three levels: which apps it can reach, which actions it can take without asking, and a set of safety checks you cannot switch off. Set the first two deliberately before you hand over anything important.
The Custom rules screen lets you assign one of four behaviors to a supported action:
Custom rule behaviors for OpenAI Dots, per OpenAI's help center
Some limits hold no matter what your rules say. Changing a password or transferring money always requires you to take over. Permanently deleting data or installing software may need approval each time. Purchases with a card saved on a merchant's website need your approval, which can be given in advance for a specific purchase.
On top of your rules, OpenAI runs a check it calls auto-review. Before certain actions, such as sending an email, auto-review compares the recipient and content against your instructions and safety requirements. It can block the action, ask you, or stop the dot. Separate safety monitoring can pause a dot that behaves in a potentially harmful way.
OpenAI says these protections reduce the risk of prompt injection without eliminating it, and that dots can still make mistakes. GPT-6 Astra, the model behind dots, is also the first OpenAI model rated Critical for cybersecurity capability under its Preparedness Framework. OpenAI explains that rating in its Path to Astra post. It is worth knowing before you give a dot broad access.
You can follow everything in Activity View on desktop, redirect the dot mid-task, or pause it from the menu on its profile.
What OpenAI Dots can't do yet
Dots launched with real gaps. Here are the limits OpenAI has confirmed in its own documentation, as of October 6, 2026.
Confirmed OpenAI Dots limits, from OpenAI's getting started guide and privacy and safety FAQ
None of these is a dealbreaker on its own. Together, they point to a product built first for desk work inside ChatGPT and Slack, not for errands that need a phone line or an inbox of its own.
What happens to data from your connected apps
Disconnecting an app stops new access, but it does not delete what your dot already learned from that app. To remove that information, you have to delete the dot itself. Decide what to connect with that in mind, especially if the app holds customer data.
Here is how retention works, per OpenAI's privacy and safety FAQ:
- Turning off ChatGPT Memory: stops memory sharing between ChatGPT and your dot, but does not delete what the dot already received.
- Deleting your dot: erases the dot's own context. Files, Codex threads, and ChatGPT conversations it created are stored separately and stay.
- Credentials and screenshots: are not kept in the dot's context. For supported sign-ins, you type your password into a secure form the model never sees.
- Training on personal plans: your "Improve the model for everyone" setting controls whether your dot's conversations and work can be used. That can include data from connected apps used in a conversation.
- Training on Business, Enterprise, and Edu: OpenAI does not train on workspace data by default.
- Human review: may still happen in limited cases, such as safety issues, even with model improvement turned off.
Your dot's proactive research and private notes are not used for training directly. They can be used if that information enters an eligible conversation.
The practical move for a business owner is to start small. Connect low-risk apps first, watch your dot for a week or two, and add sensitive systems once you trust its habits. Reversing a connection later will not undo what it already absorbed.
OpenAI Dots pricing and availability
There is no separate price for a dot. Your first dot is included at no extra cost in ChatGPT Pro and Business Premium. OpenAI says both plans include an allowance for deeper work, with extended limits for the first month after launch. Check OpenAI's current plan documentation for limits after that.
OpenAI Dots availability by ChatGPT plan, pricing as of October 2026, from OpenAI's pricing page and help center
OpenAI's pricing page shows prices in local currency, so check it from your own region. Our ChatGPT pricing guide breaks down every tier. If you are deciding between the two individual plans, see ChatGPT Plus vs Pro.
Two cost details are easy to miss. Talking to your dot does not use up ChatGPT limits, but Codex and ChatGPT Work tasks it starts do. And OpenAI plans to sell more dots and higher speed or monthly capacity per dot later, without published pricing yet.
Specialist dots and Microsoft Agent 365
Specialist dots are a separate, enterprise-only preview. Your personal dot works on your behalf. A specialist dot takes on one defined job inside a company, with its own identity, credentials, and access to the systems that job needs.
Internal OpenAI tests have covered procurement, invoice processing, email marketing, customer support, and commercial contracting. OpenAI describes the current program as focused enterprise pilots, where its engineers help each company define the dot's responsibilities, tools, and approval steps.
Microsoft and OpenAI are also working to bring specialist dots under the governance and security controls in Agent 365. For most small businesses, this is a signal of where dots are heading rather than something to plan around today.
OpenAI Dots vs Meta Muse
Dots arrived three weeks after Meta launched Muse, its own personal agent, on September 8. Both run on a dedicated cloud machine and keep working in the background. The difference is emphasis. Meta's launch examples center on personal goals and errands, while OpenAI's center on desk work inside ChatGPT and workplace chat tools. Meta has since announced Muse for Small Business, so that gap may narrow.
OpenAI Dots vs Meta Muse at launch, based on OpenAI's and Meta's own announcements
If most of your working day already runs through ChatGPT and Slack, a dot fits with less change. If you live in WhatsApp and need help with personal errands in the US, Muse is aimed at you.
When a dot isn't enough: running your apps vs owning your software
A dot operates software. It does not give you software. Every task it handles runs through an app you have connected. So the real question is whether those apps exist and fit the way you work.
For many small businesses, they do not. Bookings live in WhatsApp threads, and client records are scattered across spreadsheets. Point a dot at that setup and it will faithfully keep a messy process running, because there is no system underneath for it to work with.
The fix is usually a small, purpose-built app: a client portal, a booking system, or an inventory tracker shaped around your workflow. ChatGPT can help design one and generate its code, as we cover in ChatGPT app building. But code alone is not a working product. Your customers need something hosted, with a database behind it, secure logins, and a way to pay.
That is the job Emergent does. You describe the software your business needs in plain language, and Emergent builds a full-stack app with the backend, database, authentication, and deployment included. You own the code. Our guide to building your own CRM walks through what that looks like in practice.
Once the app exists, an agent has something real to work with. The two are not either-or.
Emergent's own always-on agent, Wingman, handles that second half. It works inside WhatsApp, Telegram, and iMessage, scheduling meetings, running sales outreach, and managing your inbox. That makes it a practical option if a dot isn't available on your plan or in your region, or if your business already runs on those chats. Apps you build on Emergent integrate with Wingman, so the software and the agent that operates it work as one system.
Should you set up an OpenAI Dot?
If you already pay for ChatGPT Pro or Business Premium, yes. Your first dot costs nothing extra, and the first month comes with extended limits. Set it up on desktop, connect two or three low-risk apps, and give it one recurring responsibility you can check easily.
If you are not on those plans, OpenAI Dots alone are not a reason to upgrade yet. The launch limits are real: one primary dot, no email address of its own, no outbound calls, and no Pro access in the EEA, Switzerland, or UK. Watch how pricing settles once the first month ends.
Either way, remember what a dot depends on. It can only run the systems you already have. If the software your business needs does not exist yet, build that first.
Most AI app builders stop at prototypes. Emergent creates production-ready apps you can actually launch.
- Production-ready apps
- Web & mobile apps
- Deploy in minutes
