OpenAI Dots are always-on agents inside ChatGPT. Each dot runs on GPT-6 Astra, works from its own cloud computer, connects to more than 4,000 apps through ChatGPT plugins, and keeps working toward a goal after you close the tab. OpenAI announced them at DevDay on September 29, 2026, and is rolling them out to Pro and Business Premium plans first.
Our view is that Dots are the first mainstream product that treats an agent as a standing colleague rather than a chat session. That makes them useful for recurring, reviewable work. It also moves the hard problems from prompting to permissions, review volume, and audit.
Sep 29
Announced at DevDay 2026
4,000+
Apps through plugins
1
Dot included on Pro and Business Premium
No API
ChatGPT, Slack, and Teams only
What are OpenAI Dots?
A dot is a persistent agent attached to your ChatGPT account. OpenAI describes dots as extensions of the user rather than conversational assistants. Each one draws on ChatGPT memory, uses Codex and connected tools, and learns from your feedback over time.
The difference from earlier ChatGPT agents is continuity. A normal agent run starts when you send a prompt and ends when it answers. A dot holds a standing goal, watches for new inputs, and picks the work back up when something changes.
OpenAI's launch examples show the intended shape of that work:
- Watch customer feedback for recurring requests, build and test small fixes, and return pull requests with a video of the change
- Revise launch materials when project scope changes
- Rerun a scientist's analyses as new data arrives
- Update enterprise sales proposals as requirements shift
- Turn interview transcripts into clips and show notes
You reach your dot through ChatGPT on web, desktop, and mobile, plus Slack and Microsoft Teams. OpenAI says texting is coming as a limited beta for Pro users in the US. You create a dot from ChatGPT on desktop or the web. After setup, you can also work with it from the mobile app.
How a dot works
A dot combines four parts: a model, a computer, a set of connected apps, and a policy layer that decides which actions run on their own.
The dot action loop
Goal and trigger
Your request, new feedback, new data, or a schedule
Dot on GPT-6 Astra
Plans with ChatGPT memory and past feedback
Custom Rules and auto-review
Allow, require approval, or block each action
Cloud computer and apps
Codex and plugin calls you can watch
Actions that need approval come back to you in ChatGPT, Slack, or Teams. Some tasks, such as changing a password, always stay with the user.
Two controls matter most. Custom Rules let you allow a specific action, require approval for it, or block it. Auto-review checks actions that affect your accounts or share information against your instructions before they run. Background research is limited to read-only tools.
You can open a dot's cloud computer while it works and watch what it is doing. That is useful for spot checks. It is not the same as an audit log you can query a week later, and OpenAI's launch materials do not describe one for individual dots.
Who gets Dots and what they cost
The rollout has more conditions than the headline suggests:
| Plan | Access | Cost |
|---|---|---|
| Pro | Rolling out in eligible markets | First dot included |
| Business Premium | Rolling out in eligible markets | First dot included |
| Enterprise, Edu, Healthcare | Beta, off by default, admin must enable | Not yet published |
| Additional dots | Announced | Pricing coming later |
Conversations with a dot do not count toward ChatGPT usage limits. Tasks a dot starts in Codex or ChatGPT Work count normally. A dot that opens pull requests all night can therefore spend the same Codex allowance your engineers rely on during the day.
For organizations, OpenAI previewed specialist dots that have their own identity and credentials, plus an integration with Microsoft Agent 365 for governance. Both are previews. Plan around the single, user-owned dot that ships today.
A starting policy for your first dot
Custom Rules are the part teams will get wrong first. A dot that can do everything will produce more output than anyone can review, and a dot that must ask about everything stops being always-on. We start from this split and loosen it only after a week of reviewed results.
| Rule | Actions | Why |
|---|---|---|
| Allow | Read issues, docs, and dashboards; draft in a scratch folder; open draft pull requests | Reversible and easy to review |
| Require approval | Post in shared channels, comment on customer tickets, edit shared docs | Visible to other people |
| Block | Merge, deploy, send email, change billing, delete files, invite users | Hard to undo or outward-facing |
Give the dot one written goal with a finish condition, the same way you would brief a contractor. For example: "Each weekday, group new support tickets that mention checkout. When three or more share a cause, open a draft PR with a failing test and a fix. Do not touch payment code." The last sentence matters as much as the first.
Dots versus building your own agent
There is no Dots API. Developers who want the pattern inside a product use the building blocks OpenAI shipped at the same DevDay: the Agents API with hosted computer use, and support for the proposed MCP Events specification for webhook-driven triggers.
| Question | OpenAI Dots | Your own agent |
|---|---|---|
| Who uses it | One ChatGPT user | Your customers or teams |
| Where it runs | OpenAI cloud computer | Your choice, including your own cloud |
| Model choice | GPT-6 Astra | Any, including mixed models |
| Audit trail | Watch live; log export not described at launch | Whatever you build and retain |
| Cost model | Seat plus shared usage limits | Per-token API billing |
| Time to first result | Minutes | Weeks |
Dots win for personal and small-team work where the person who owns the dot also reviews its output. A custom agent wins when the agent serves other people, touches regulated data, or needs a durable record of what it did. We covered how to judge which workflows justify GPT-6 Astra's cost in our GPT-6 Astra workflow guide.
The honest limits of Dots
Review volume is the real bottleneck. An agent that works overnight produces a morning queue. If nobody has time to review five draft PRs a day, the dot is creating debt, not saving time. OpenAI's own guidance is that dots can still make mistakes, so consequential work needs review.
Safeguards are new. Custom Rules and auto-review are sensible, but they are a first release. We treat them as a first attempt to bound the risk, not a proven fix. Treat any plugin a dot can write to as reachable by a prompt injection hidden in content the dot reads. Our tool use checklist for AI agents covers how to scope write access.
Memory cuts both ways. A dot that learns your preferences also learns your shortcuts. If you approve a sloppy result once, expect more of it.
Lock-in is real. A dot's learned context lives in ChatGPT, and OpenAI has not described a way to move it to another vendor or to your own agent.
When Dots are not worth using
Skip Dots when the work has no reviewable output, such as "keep an eye on things." Skip them when the work involves customer data your contracts keep inside your cloud. Skip them for one-off tasks, where a normal ChatGPT or Codex session is cheaper and easier to inspect. And skip them if your team already struggles to review human pull requests.
Use one when a recurring task has a clear trigger, a reversible output, and a named person who reviews it. Start with a single dot on a single workflow for two weeks, count the results you accepted without edits, and decide from that number.
FAQ
What are OpenAI Dots?
Dots are always-on agents in ChatGPT, announced on September 29, 2026. Each dot runs on GPT-6 Astra, has its own cloud computer, connects to more than 4,000 apps through plugins, and learns from feedback. Unlike a normal chat, a dot keeps working toward a standing goal in the background and returns results for your review.
How much do OpenAI Dots cost?
The first dot is included at no extra cost on ChatGPT Pro and Business Premium plans in eligible markets. Enterprise, Edu, and Healthcare workspaces get a beta that an admin must enable. OpenAI says pricing for additional dots is coming later. Tasks a dot starts in Codex or ChatGPT Work still count toward your usage limits.
Is there an API for OpenAI Dots?
No. Dots are available only through ChatGPT, Slack, and Microsoft Teams, with texting coming as a limited beta for US Pro users. Developers who want similar behavior inside their own products can use OpenAI's Agents API with computer use, which launched at the same DevDay, and handle triggers, approvals, and logging in their own code.