OpenAI Dots are always-on AI agents inside ChatGPT that keep working after you close the chat window. Each Dot runs on GPT-6 Astra, gets its own cloud computer, connects to more than 4,000 apps, and can start and finish multi-step work on its own — investigating a bug mentioned in Slack, turning a finished design into a working app, or noticing an unsent invoice and preparing it for approval. Your first Dot is included with ChatGPT Pro (from $100/month) or Business Premium at no extra cost, but the chats are the free part. The tasks your Dot starts bill against your usual usage allowances, so the real cost depends on how much work you let it do.
If you run a small business, Dots are worth understanding now, not because they’re magic, but because they change the job from “prompting AI” to “managing AI.” You set a goal, the Dot works, it reports back, and you make the decisions. Done right, that clears the recurring work that never quite gets automated. Done carelessly, it hands an agent the keys to your inbox, your files, and your customer-facing channels with rules you never wrote down. This guide covers setup, the real cost model, a safety checklist to run before you leave a Dot unattended, and how Dots compare with Meta Muse.
What a Dot Actually Is (and What It Isn’t)
A Dot is a persistent personal agent, not a smarter chatbot. A normal ChatGPT conversation answers your prompt and waits. A Dot receives a goal, works through the steps, uses the apps and tools you’ve authorized, and keeps going until it finishes or needs a human decision. It remembers context across days through its own notes plus ChatGPT memory, and it can carry that context across ChatGPT, Slack, and Teams.
Here is what that looks like in practice, using OpenAI’s own examples:
- It watches while you don’t. A bug gets mentioned in Slack. Your Dot investigates it, pulls the relevant context, and comes back with findings instead of waiting for you to ask.
- It finishes multi-step work. A finished design becomes a working app. A research request becomes a finished document, spreadsheet, or deck.
- It works between conversations. When you’re not delegating, a Dot does “proactive research” — reading permitted apps and taking private notes — so it can surface things you missed, like that unsent invoice.
What it isn’t: an employee, a replacement for your judgment, or an unlimited worker. You start with one primary Dot. OpenAI has talked about teams of Dots and specialist Dots for businesses, but at launch, one is what you get. There is no public API for Dots themselves, and the agent still stumbles on things that need human hands — early testers found it tripping over security checks that required manual mouse clicks, like confirming a bank-account-based discount.
The honest way to think about it: a Dot is a tireless junior operator with a good memory and no common sense of its own. Your job is to give it narrow responsibilities, clear rules, and a short leash on anything a customer will ever see.
Who Gets a Dot, and What It Really Costs
Dots launched at OpenAI’s DevDay on September 29, 2026 (OpenAI’s Dots announcement), and the rollout to eligible Pro and Business Premium subscribers began the same day. Access is gradual, so eligibility doesn’t guarantee instant access — some accounts wait days.
| Plan | Price | Dot included? | Region |
|---|---|---|---|
| Free | $0 | No | — |
| Go | $8/month | No | — |
| Plus | $20/month | No | — |
| Pro 100 | $100/month | Yes, first Dot | Outside EEA, Switzerland, UK |
| Pro 200 | $200/month | Yes, first Dot | Outside EEA, Switzerland, UK |
| Pro 500 | $500/month | Yes, first Dot | Outside EEA, Switzerland, UK |
| Business Standard | $20/user/month (annual) | No | — |
| Business Premium | $100/user/month (annual) or $125 monthly | Yes, first Dot | All supported regions |
| Enterprise / Edu / Healthcare | Custom | Beta, off by default, admin enables it | — |
A few things in that table matter more than the headline prices:
Your first Dot is included — but “included” has a catch. Conversations with your Dot don’t count against ChatGPT usage limits. The tasks it starts in ChatGPT Work or Codex do count, at the usual rates. OpenAI extended deeper-work limits for the first month, but always-on access does not mean unlimited tasks. If your Dot runs a dozen research-and-build jobs a day, you will burn through your allowance fast.
The Pro 200 allowance is shrinking. On October 30, 2026, the ChatGPT Pro ($200/month) Codex + Work allowance drops from 20× Plus to 10× Plus. That halving hits Dot users directly, because Dot jobs bill against those same meters. If you’re on Pro 200 and your Dot is a heavy worker, that deadline is your real budget conversation — not the Dot itself, which costs nothing extra.
You need to be 18+, and Pro users in the EEA, Switzerland, and the UK are excluded at launch. Business Premium has no such regional carve-out. Free, Go, Plus, and Business Standard users don’t get a Dot at all, and OpenAI hasn’t given a date for expanding access.
More Dots will cost something, eventually. OpenAI has said you’ll be able to add more Dots or buy more capacity later, with prices still unannounced. Budget for one Dot per person for now, and treat anything beyond that as a future line item.
So the real cost math for a small business is simple: the Dot itself is $0 on top of a plan you’re either already paying for or upgrading to. The upgrade is the cost. If you’re on Plus at $20/month and you move to Pro 100 at $100/month for the Dot, you’re paying $80/month for an agent whose jobs still draw down your usage meter. That’s a fair price if it genuinely clears recurring work — and a waste if you only needed a chatbot with a better memory.
Setting Up Your First Dot
Setup takes a few minutes, but the five minutes you spend on rules afterward matter more than the setup itself.
- Create your Dot on desktop. Open the ChatGPT desktop app or ChatGPT in a desktop browser. You cannot create a Dot on mobile — you can only message it from the mobile app after setup. Mobile web is unsupported for creation.
- Name it and describe its job. Give it a name and a plain-language description of what it’s for. “You track my open invoices, watch my support inbox for urgent messages, and prepare a morning brief” beats “help me with my business.” Specific beats broad with agents, because broad is where scope creep lives.
- Connect only the apps the job needs. Each connected app becomes material your Dot may read in the background. Connect Slack if it watches Slack. Don’t connect your email, your files, your calendar, and your CRM on day one “just in case.” You can add access later; you can’t un-teach it what it already read.
- Write your Custom Rules before the first task. These are the policy layer (more on them below). Set them before you give the Dot real work, not after it surprises you.
- Start with one small, visible job. Have it prepare tomorrow’s morning brief or summarize one thread. Watch how it reports back in Activity View. Then widen its responsibilities one job at a time.
That last step is the one most people skip, and it’s the one that prevents most disasters. A Dot that has done five small jobs well is a Dot you can trust with a sixth. A Dot handed your whole operation on day one is a gamble.
One Dot Per Function: The Small-Business Playbook
You only get one Dot, so you can’t literally run one per function yet. But you can still organize its work by function, with a separate standing instruction for each. Here are the three functions where small businesses get the most value, in order of safety:
1. SEO and content research (safest, start here)
Give your Dot a standing job: every weekday morning, research one topic in your niche, pull the current facts, and draft a brief with sources linked. This is read-mostly work with an obvious output you review before anything goes public. It replaces the “open twelve tabs” morning routine, and the failure mode is a mediocre draft, not a sent email.
Be specific about sources. Tell it which publications, competitor blogs, and official docs to check, and require a link next to every factual claim. If it can’t verify something, it should say so, not fill the gap.
2. Campaign and operations monitoring
A Dot can watch for the things that go stale: an ad campaign whose spend drifted, a landing page with a broken form, a listing whose price no longer matches your site, a competitor who changed their offer. The output is an alert, not an action. You decide what to do about it.
This is where Dots earn their keep for operators. Humans are bad at checking the same five dashboards every day. Agents are good at it — as long as “checking” doesn’t quietly become “changing.”
3. Inbox and invoice triage (valuable, but leash it)
A Dot can notice the unsent invoice, flag the client who hasn’t replied in a week, and draft — draft, not send — the follow-up. It can also prepare your morning brief from overnight messages.
This is the function where the safety checklist below matters most, because inbox access is the most sensitive access you’ll grant. Start with read-only summaries and drafted replies. Promote it to sending only after a week of drafts you’d have sent unchanged.
What not to hand a Dot in its first month: anything that spends money, sends messages to customers, changes published content, or touches credentials. Those are jobs for a Dot with a track record, explicit rules, and your approval step still in the loop.
The Safety Checklist Before You Leave a Dot Unattended
This is the section to actually follow. Dots are designed with real guardrails, but guardrails only work if you configure them. Run through this list before your Dot’s first unsupervised night.
1. Set Custom Rules — and know their limits. In Settings → Personalization → Custom Rules (under Permissions), you set four behaviors for each kind of action: act without asking, act when told, ask first, or hand off to you. Put anything a customer sees — sending messages, publishing, replying to reviews — on “ask first” or “hand off.” But know what rules can’t do: they can’t grant app or computer access, can’t override built-in safety, can’t remove required confirmations, and can’t relax auto-review or proactive-research restrictions. A workspace admin can also disable them. Rules are a policy layer, not a master key.
2. Understand auto-review. A separate check compares every consequential action against your instructions, permissions, rules, and OpenAI’s safety requirements, then decides: run, ask, hand off, or block. You cannot switch auto-review off. Fixed classes of actions — password changes, money transfers — always need you. Deleting data or installing software may need per-action approval. Purchases need approval unless you specifically pre-authorized them. Design your workflows assuming some steps will stop for a human, because they will.
3. Know what proactive research can and can’t do. While idle, your Dot reads permitted apps and takes private notes. That background research is read-only by design: its research tools cannot message people, change your content, or drive a browser or computer. Acting on a finding goes through normal approvals. Don’t assume “it can act when I’m not looking” for anything consequential — by design, it can’t.
4. Use Activity View. This is your window into what the Dot is doing: ongoing and delegated tasks, the steps it has taken, and a live view of its cloud computer. Check it daily for the first two weeks. If you can’t explain what your Dot did yesterday, its job description is too vague.
5. Keep task chains short. OpenAI’s own testing showed scope problems growing as task chains got longer — the agent drifting from what you actually asked. Prefer many small, verified steps over one long autonomous chain, and re-check the state of the world between steps. “Research five competitors and draft a comparison” is one reasonable job. “Research competitors, rebuild my pricing page, email my list about it, and update my ads” is a chain waiting to go sideways.
6. Treat memory as permanent. Your Dot’s memory comes from three sources — working context, ChatGPT memory, and its own notes — and you cannot view, correct, or delete individual Dot memories. Deleting the Dot is the only reset, and disconnecting an app stops future access without erasing what it already learned. Nothing that touches secrets, regulated data, or anything you cannot re-derive belongs in its reach. Ever.
7. Keep your computer separate. Your Dot runs on its own isolated cloud computer. Your own machine stays separate unless you explicitly connect it through the desktop app, and that connection is off by default. Leave it off until you have a specific job that needs it.
8. Treat viral anecdotes with skepticism. At least one widely shared claim about a Dot contacting officials on a user’s behalf has not been confirmed by OpenAI. Dots are new, the discourse is loud, and the gap between “it could do that” and “it did that” is where most agent horror stories live. Verify before you panic — and before you repeat one.
One more honest caveat: Dots can make mistakes. OpenAI says so, and every early review agrees. Consequential work should still be reviewed. The Dot that drafts your invoice follow-up is a gift; the Dot whose draft you send without reading is a liability.
Dots vs Meta Muse: Which Agent Fits Which Business?
Dots aren’t the only always-on agent on the market. Meta’s Muse launched September 8, 2026, and it’s the most direct alternative — a personal AI agent that also works through its own cloud computer and keeps going after you close the app. Here’s how they compare for a small business:
| OpenAI Dots | Meta Muse | |
|---|---|---|
| Price | First Dot included in Pro ($100–$500/mo) or Business Premium ($100/user/mo annual) | Free tier (reported 100M tokens/week), Power $20/mo, Maximum $100/mo |
| Model | GPT-6 Astra | Muse Spark |
| Availability | Pro excluded in EEA, Switzerland, UK; Business Premium in all supported regions | US and Canada |
| Where you reach it | ChatGPT, Slack, Teams, voice; texting in limited US Pro beta | App, web at muse.ai, WhatsApp, Mac |
| App connections | 4,000+ via plugins | Email, calendar, payments, shopping, travel, smart home |
| Strength | Coding, research, multi-step work inside ChatGPT | Everyday errands, shopping, scheduling, household admin |
| Commerce | Via connected apps | Built in: Shopify, Shop Pay, PayPal, Stripe Link purchase protections |
The choice isn’t really “which agent is better.” It’s “which ecosystem do you already live in, and what’s the job?”
- Pick Dots if your work lives in ChatGPT, Slack, or Teams, if coding or technical research is a real part of your week, or if you’re already paying for Pro or Business Premium. The Dot is then nearly free, and its coding chops are the differentiator.
- Pick Muse if you want the cheapest on-ramp (free is hard to beat), if your work is everyday errands and shopping rather than code, or if your team coordinates over WhatsApp. For a US-based solo operator who mostly needs admin help, Muse’s free tier is the rational first try.
- Use neither for money movement until each has a track record with you. Both agents ask before sensitive actions by design — keep it that way.
One caution that applies to both: the free or cheap tier is not the whole price. With Dots, the cost is the plan upgrade plus usage. With Muse, the cost is attention — it’s a Meta product, and its chats feed ad personalization. Read the privacy terms of whichever you choose like the business decision it is.
Where Dots Will Disappoint You
I want to be straight about the limits, because the launch hype skips them:
- The first month is a honeymoon. OpenAI extended deeper-work limits for the first month. Nobody has published what the steady-state allowance looks like after that. Your cost math in month two could differ from month one, and you should budget for it.
- Memory is opaque. You can’t audit what your Dot remembers, line by line. If it picks up a wrong fact about your business — a stale price, a misremembered policy — it may keep acting on it. Correcting that means deleting the Dot and starting over.
- It’s one Dot. You can’t yet split responsibilities across a team of Dots with separate identities. Everything runs through one agent with one set of notes, which is fine for a solo operator and limiting for a team.
- The UX is still rough. Early reviewers note that just getting to Dots means rummaging around in the ChatGPT app’s sidebar, and that Dots, Chat, Work, and Codex overlap in confusing ways. Expect some fumbling.
- It is not a strategist. Dots execute well-defined jobs. “Grow my business” is not a well-defined job. The quality of what you get out is capped by the quality of the goals and rules you put in.
None of that makes Dots a bad product. It makes it a first-generation product. The businesses that get value from it in 2026 will be the ones that treat it like a promising new hire: narrow job description, clear rules, daily check-ins, and no unsupervised access to the safe.
The Bottom Line
OpenAI Dots turn ChatGPT from a tool you operate into an agent you manage. For a small business, the smart move is to start with one read-mostly job — research, monitoring, or a morning brief — write strict Custom Rules before the first real task, keep anything customer-facing on “ask first,” and let the Dot earn wider access one good week at a time. If you’re already on Pro or Business Premium, your first Dot costs nothing extra to try. If you’re not, weigh the upgrade against Meta Muse’s free tier and against the honest question of whether you have recurring multi-step work that actually needs an agent — or just a better prompt.
Watch the video below for OpenAI’s own walkthrough of how Dots work and the controls around them.
Frequently Asked Questions
What are OpenAI Dots?
Dots are always-on AI agents inside ChatGPT, launched September 29, 2026. Each Dot runs on GPT-6 Astra, has its own cloud computer, connects to more than 4,000 apps, and works on goals between conversations instead of waiting for your next prompt.
How much does an OpenAI Dot cost?
Your first Dot is included in ChatGPT Pro ($100, $200, or $500/month) or Business Premium ($100/user/month billed annually, $125 monthly) at no extra cost. Chats with your Dot don’t count against usage limits, but the Work or Codex tasks it starts do — so heavy use still costs usage allowance.
Can I get a Dot on ChatGPT Plus or Free?
No, not at launch. Dots are only on Pro and Business Premium, plus an admin-enabled Enterprise beta. OpenAI has said it plans to expand access, with no date given.
How do I set up my first Dot?
Create it in the ChatGPT desktop app or a desktop browser — not on mobile. Name it, describe its job in plain language, connect only the apps that job needs, write your Custom Rules, and start with one small visible task before widening its responsibilities.
Are Dots safe to leave running unattended?
Reasonably safe if configured: proactive research is read-only, auto-review can’t be switched off, and password changes or money transfers always need a human. But memory can’t be selectively edited, and scope drift grows with long task chains — so keep customer-facing actions on “ask first” and review Activity View daily at first.
What’s the difference between Dots and Meta Muse?
Dots run on GPT-6 Astra inside the ChatGPT/Slack/Teams ecosystem and come with Pro or Business Premium plans. Muse runs on Muse Spark, is free to start (US and Canada), and lives in its own app, the web, WhatsApp, and Mac. Dots lean toward coding and research; Muse leans toward everyday errands and shopping.
Sources: OpenAI Dots pricing breakdown · Digital Trends on the Dots launch · Gemini 4, GPT-6 Astra and Claude flagship pricing comparison · Meta Muse Wonder frontier model pricing and benchmarks