OpenAI dots are always-on agents that live inside ChatGPT. Each dot has its own cloud computer, reads the apps you connect, and works toward goals you set while you are away, pausing for approval on anything risky. As of November 2026 the first dot comes with ChatGPT Pro (from $100/month) or Business Premium. This review is based on OpenAI's documentation, launch coverage and published hands-on reports; we have not run our own tests.
What is an OpenAI dot?
OpenAI announced dots at DevDay in San Francisco on September 29, 2026, with Sam Altman calling them "remarkably capable, always-on agents built to handle really anything you can think of" (CNBC). The rollout started the same week for Pro and Business Premium users in eligible markets.
A dot is not a chat mode. It is a persistent agent you name and customize, with a cartoonish floating avatar that TechCrunch compared to Meta's Muse persona (TechCrunch). You start with one primary dot. OpenAI has said "teams of dots" are planned, and businesses can pilot role-specific "specialist dots" with their own organizational identity.
If you are new to the category, our explainer on what a personal AI agent is covers the basics.
What can a dot actually do?
The documented and demonstrated tasks fall into a few groups.
Background work toward goals
Dots keep working after you close the laptop (VentureBeat). CBS listed scheduling meetings, booking travel, budgeting, debugging systems and writing code as target uses. OpenAI's own example is a dot that noticed an unsent invoice and prepared it for the user to approve.
Watching channels
A dot can monitor Slack channels you name and alert you when conditions you set are met (learn.chatgpt.com). It works inside ChatGPT, Slack and Microsoft Teams, and since October 9 it can start work in Codex and edit Scheduled Tasks in ChatGPT Work.
Keynote demos
At DevDay, a dot called "Alfred" published a website after a phone call, "Felipe" edited a dashboard by voice and flagged a calendar clash, and "Dottie" built a test music app and replied in Slack. Demos are not the same as everyday reliability, so treat them as the ceiling rather than the norm.
What it cannot do yet
The marketing line "calls and writes for you" goes further than the docs. You can call your dot, but dot-initiated calls are "planned for after launch", it can text only you, and it messages other people in Slack and Teams only when you tell it to (learn.chatgpt.com). It cannot phone a restaurant or a customer service line.
Spec sheet
- Cheapest plan with a dot (Pro 100)
- $100/mo
- Connected apps
- 4,000+
- Announced at DevDay
- Sep 29, 2026
- Personal Pro availability (Nov 2026)
- No UK/EEA
| Item | OpenAI dots (as of November 2026) |
|---|---|
| Launched | September 29, 2026 (DevDay) |
| Model | GPT-6 Astra; GPT-6.1 Terra added November 3 for parts of the work |
| Runs on | Isolated cloud Linux computer with Chrome; optional link to your own computers |
| Connectors | 4,000+ apps through OpenAI's plugin ecosystem |
| Where you talk to it | ChatGPT (iOS, Android, desktop app, desktop web), Slack, Teams, voice calls, iMessage/RCS texting beta (US) |
| Entry price | ChatGPT Pro 100 at $100/month; Business Premium |
| Free tier | No |
| Watch or take over | Yes, via its screen and Activity View |
| Personal Pro availability | Outside the UK, EEA and Switzerland; 18+ |
How does a dot work under the hood?
Each dot gets its own cloud computer, a Linux environment with Chrome that is isolated from other users. You can watch its screen and press "Take over" when it hits something only you should do (Decrypt). Connecting your own local computer is optional and off by default. Since November 9 you can connect several local computers to one dot and choose which one runs a task (OpenAI Community).
- You (ChatGPT, Slack, text)
- Your dot
- Auto-review + Custom Rules
- Cloud computer with Chrome
- Connected apps
Personal dots act under your own connected accounts. A dot has no email address of its own, which is a difference from Gemini Spark and from Grok Bot's later email feature. Our guide to how personal AI agents work explains why cloud computers matter.
Memory
A dot receives memories from ChatGPT and forms its own, including from connected apps. Memory carries across ChatGPT, texting, Slack and calls. The weak point: individual dot memories cannot be viewed, edited or deleted. Turning off ChatGPT Memory stops new sharing but does not erase what the dot already received, and disconnecting an app does not remove what it learned (OpenAI Help). The only reset is deleting or resetting the dot. We compare this with rivals in agent memory and privacy compared.
How safe is it? Approvals and limits
OpenAI layers several controls, according to its documentation and VentureBeat's launch report:
- Proactive research is read-only, enforced in code. While working on its own initiative, a dot cannot send messages, edit app content or drive a browser.
- Auto-review checks each planned action, such as an email's recipient and content, against your instructions, your rules and OpenAI's safety requirements. The dot cannot turn it off.
- Custom Rules give four behaviors per action type: act without asking, act if pre-approved, ask first, or hand off to you.
- Hard limits: password changes and money transfers always come back to you; saved-card purchases need approval; one approval does not grant ongoing permission.
- Credentials go through a secure form and are not exposed to the model.
OpenAI says its injection defenses "reduce" the risk but "do not eliminate it". Its GPT-6 Astra system card calls the model its most prompt-injection-robust so far. In the card's Gray Swan test, Astra's prompt-injection attack success rate was 8.5% across 1,810 attacks, against 27.0% for GPT-5.6 Sol (OpenAI). Crypto Briefing also reports scope-violation rates for dots, but we could not find those numbers in the published system card. One more data point: OpenAI pulled GPT-6.1 Astra in October because it "didn't quite meet the bar of staying within scope and authorization."
For a side-by-side of what each agent may do unprompted, see agent approval rules compared.
How much do dots cost?
The first dot is included with ChatGPT Pro or Business Premium. It is not available on Free, Go ($8) or Plus ($20).
| Plan | Price (as of November 2026) | Dots |
|---|---|---|
| ChatGPT Plus | $20/month | No |
| Pro 100 | $100/month | First dot included |
| Pro 200 | $200/month | First dot included |
| Pro 500 | $500/month (adds Ultrafast) | First dot included |
| Business Premium | $125/user/month monthly or $100 annual, 2-seat minimum | First dot included |
Business Premium costs $125 per user a month billed monthly or $100 billed annually (OpenAI); CNBC's $20 figure was the annual price of a Standard seat, which does not include a dot. Dot conversations do not count against ChatGPT usage limits; they draw on a separate "deeper work" allowance, extended for the first month, while tasks a dot runs in Codex or ChatGPT Work do count. Extra dots and faster speeds are planned paid add-ons, and OpenAI has not published their prices or the allowance after the first month. Full detail is on our OpenAI dots pricing page.
Where is it available?
Personal Pro access is for users 18 and older outside the EEA, Switzerland and the UK. EU Business and Enterprise workspaces got access on October 28, 2026, and Enterprise, Edu and Healthcare became generally available on November 4, with dots switched on by default from November 18 unless an admin opts out. Admins can read our Enterprise default-on guide for what to decide before then. No UK Pro date has been announced. UK readers should check personalagent.uk's dots guide, and our availability by country tracker covers the rest.
What do published reviews say?
- The Neuron (day one) found it "feels natural to use"; its dot flagged a warranty email and drafted a reply, but once sent an email before the reviewer had checked it.
- eesel gave 3/5: strong on safety design (4.5) and capability (4), weak on friction (2.5) and availability (2). eesel sells a competing product.
- Friction complaints: a shuttle booking needed several confirmations, and the dot would not fetch a 2FA code from an inbox it could read. Others called it slow or buggy, and Neoteo found online errands stalled on security checks.
- Positioning: analyst Adam Crisafulli told CBS it is "very much geared toward enterprise/professional users, not everyday consumers."
Pros
- 4,000+ app connectors and Slack/Teams presence
- You can watch the cloud computer and take over
- Read-only proactive mode and auto-review that the dot cannot disable
Cons
- Starts at $100/month; no Free or Plus access
- Memories cannot be viewed or deleted individually
- Personal Pro excludes the UK, EEA and Switzerland
- Reviewers report confirmation-heavy errands and slow speed
Who it's for, and who should skip it
Good fit: people already paying for ChatGPT Pro, and teams on Business Premium who live in Slack, Teams, email and Codex. Dots connect most directly to work tools, and you can watch and take over the computer.
Skip it if: you want a free or cheap consumer agent (Meta Muse has a free tier; see Meta Muse vs OpenAI dots), you are in the UK or EEA on a personal plan, or you need to edit the agent's memory item by item. If safety-first behavior matters more than breadth, compare Claude Duet vs OpenAI dots.
Bottom line
Based on published features and pricing, dots are the most connected of the 2026 agents, with 4,000+ apps, a visible cloud computer and solid approval rules. The costs are a $100 entry price, no memory editor, regional limits and confirmation-heavy errands. It suits professionals who already use ChatGPT heavily more than casual users.



