What Are OpenAI Dots? Always-On Agents and Memory

OpenAI introduced dots at DevDay on September 29, 2026. OpenAI writes the name in lowercase: a dot is one agent, and dots is the product. A dot is an agent inside ChatGPT that keeps working when you are not talking to it. It has its own cloud computer and browser. It connects to your apps and builds up context about you over time. Dots are rolling out to ChatGPT Pro and Business Premium users first. If you plan to hand a dot real work, its memory design matters more than its feature list.

What are OpenAI dots, in one sentence?

A dot is an always-on agent in ChatGPT, powered by GPT-6 Astra, that runs on its own cloud computer and keeps working on your goals between conversations.

That wording follows OpenAI's getting started guide. The launch post adds the selling points. OpenAI says dots learn from feedback, can work toward your goals 24/7 and reach over 4,000 apps through plugins. You give a dot a name, and its default handle is @yourname-dot.

Dots were one of more than 20 announcements in OpenAI's DevDay 2026 recap. InfoQ's recap on October 2 called them "persistent agents that can work on ongoing tasks."

How do dots differ from Codex cloud and a local coding agent?

The closest comparisons are OpenAI's own Codex cloud, which got its own DevDay update, and a coding agent on your laptop such as Claude Code. All three can run long tasks. The difference is who starts the work and where the memory lives.

Sources: dots from OpenAI's help center, Codex cloud from OpenAI's Codex cloud docs and the DevDay recap, Claude Code from Anthropic's memory docs.

The two OpenAI products also connect. A dot can create tasks in Codex cloud environments, as long as you set the environment up in Codex first.

How do dots work?

You create a dot in the ChatGPT desktop app or on desktop web. Creation is not possible on mobile, but you can message the dot from the mobile app once it exists. You can also reach it in Slack and Teams. OpenAI's help center says texting is a limited beta for Pro users in the US.

Work reaches a dot in three ways. You ask it for something. You schedule a reminder or a recurring check. Or the dot starts on its own through what OpenAI calls proactive research. Per the safety post, proactive research uses read-only tools on your connected apps and saves private notes for the dot. OpenAI says it enforces those limits in code. The research tasks cannot message people or change app content. They also cannot control a browser or computer.

When a dot wants to act, a separate system called Auto-review checks the step against your instructions and safety rules first. You set your own boundaries with Custom Rules. Each rule picks one of four behaviors: act without asking, act if pre-approved, ask first, or hand off to you. Some steps always go back to you, such as changing a password or transferring money. You can follow ongoing and delegated tasks in Activity View in the desktop app, and you can open the dot's computer at any time.

What do dots cost, and who can use them?

Dots are rolling out in ChatGPT, and access may take several days to reach your account. The getting started guide lists:

• Pro users in all markets except the European Economic Area, Switzerland and the UK
• Business Premium users in all supported ChatGPT regions
• Enterprise workspaces (including Edu and Healthcare) as a beta that an admin must turn on

Users under 18 cannot use dots yet.

Your first dot is included in a Pro or Business Premium plan at no extra cost. The plan also includes an allowance for deeper work, with extended limits for the first month after launch. OpenAI has not published that allowance as a number. It says you will later be able to add more dots and buy more speed or monthly work per dot. 9to5Google also found no price for extra dots.

Conversations with your dot do not count toward ChatGPT usage limits, which helps heavy users. Tasks the dot starts in Codex or in ChatGPT Work, OpenAI's desktop workspace for longer tasks, do count.

How do dots handle memory?

A dot has two memory layers.

The first is shared with ChatGPT. OpenAI's privacy FAQ says a dot receives memories and recent context from ChatGPT. Conversations with the dot can also feed back into ChatGPT Memory. Turning off Memory in ChatGPT stops the sharing. It does not delete what the dot already received.

The second is the dot's own context. OpenAI's FAQ says the dot can retain context from your conversations and plugins for as long as you keep the dot. Through proactive research, it can also read permitted connected sources before you ask and save private notes. The FAQ describes these controls:

• You currently cannot see or edit individual dot memories. That includes details that came in through plugins. You can ask the dot to help correct a mistake.
• Disconnecting an app stops new access. What the dot already learned from it stays.
• The only way to clear the dot's own context is to delete the dot. Reset deletes its saved memories and scheduled tasks along with its conversations.
• Files the dot created are stored separately and survive the reset. So do its Codex threads and ChatGPT conversations.
• The context does not keep credentials. Images and screenshots are not kept either.

OpenAI says content is encrypted at rest and in transit. On personal plans, your "Improve the model for everyone" setting decides whether dot conversations and work may be used for training. OpenAI says it does not train directly on proactive research notes. A note that enters an eligible conversation may still be used, depending on that setting. By default, OpenAI does not train on workspace content from Business and Enterprise plans, including Edu.

What should you read in the fine print?

Most of what you can read about dots today comes from OpenAI. The capability claims are OpenAI's own: learning your preferences, working 24/7, sometimes finishing work "before you even think to ask." The launch examples are internal or from early testers. We found no independent test of how well a dot performs on real work. The InfoQ and 9to5Google pieces restate the announcement and add no measurements.

OpenAI is direct about the limits. Its privacy FAQ says protections against prompt injection "help reduce the risk" but do not eliminate it. The launch post tells users to always review consequential work, and the FAQ notes that some actions cannot be undone.

A dot builds context from your apps in the background, and you cannot audit that context entry by entry. If a dot picks up something wrong or sensitive, you can ask it to correct course. The only documented way to clear it is to delete the dot and start over. The OpenAI pages we read do not say how much a dot can remember or whether you can export what it knows.

The bottom line

If you have Business Premium, or Pro in a market where dots are available, create a dot. Give it bounded work that you can review. Recurring research and draft preparation are good first jobs, in our view. Connect apps one at a time, because anything the dot reads stays in its context until you delete it. Keep team knowledge that must be correct and portable somewhere you can read and edit.

How do you keep agent memory you can inspect and move?

A dot's memory is useful and opaque at the same time. It learns from your apps in the background. You cannot currently view or fix a single entry, and the only documented reset is deleting the whole agent. Its context also stays inside ChatGPT. When you switch to Claude Code or Cursor, none of it comes along. Claude Code and Codex each keep their own memory if you use them too. The dot's is the one you cannot read.

memU keeps memory as a shared wiki of readable Markdown files. A scheduled background task slices new session logs from each host into jobs. The agent decides what is worth keeping and writes it as memory or skill Markdown. memU itself makes no LLM or chat calls: it stores, embeds and retrieves the Markdown the agent prepared. A standing instruction in CLAUDE.md or AGENTS.md tells the agent to retrieve before it answers. All hosts share one backend, so what Codex learned is available to Claude Code. Because the memory is Markdown, you can open a wrong entry and fix it. Supported hosts include Claude Code and Codex. memU does not support ChatGPT Chat mode, so it cannot connect to a dot.

To try the hosted version, which is free and syncs across devices, get an API key from memu.so and send your agent one message:

Read https://memu.pro/SKILL.md, follow its instructions to install and configure memU, API Key is YOUR_MEMU_API_KEY.

The self-hosted setup and the support status for each agent are in the memU GitHub repository.

FAQ

Are OpenAI dots free?

Your first dot is included in ChatGPT Pro and Business Premium at no extra cost, with an allowance for deeper work. OpenAI has not published prices for extra dots or for more speed or work per dot.

Are dots available in the EU or UK?

Not on Pro yet. OpenAI's help center excludes Pro users in the EEA as well as Switzerland and the UK. Business Premium users get dots in all supported ChatGPT regions.

Can I see or delete what my dot remembers?

Not entry by entry. OpenAI says you cannot view or change individual dot memories. Deleting the dot clears its own context, while memories it shared with ChatGPT are managed in ChatGPT Memory settings.

Can a dot use my own computer?

Yes, if you connect it. Local computer access is optional and starts turned off. You connect a computer through the ChatGPT desktop app and can revoke access at any time.

What model powers OpenAI dots?

OpenAI's launch post and help center both name GPT-6 Astra as the model.