What Is OpenAI Dots? Always-On Agents, Explained

Published October 2, 2026 · by the AQ team

OpenAI Dots are always-on personal agents in ChatGPT, announced at DevDay on September 29, 2026. Each dot runs on GPT-6 Astra, OpenAI's flagship model, and is provisioned with its own cloud computer and browser, so it keeps working toward a goal between conversations and after your laptop is closed, then brings results back for your review. You reach the same dot from ChatGPT, Slack, and Microsoft Teams, and it can connect to more than 4,000 apps through OpenAI's plugin ecosystem. For coding, a dot is a coordinator rather than the coder: it creates and steers Codex tasks against your repositories and reports back with the result. This guide explains what OpenAI shipped, what "an agent with its own cloud computer" actually means for coding work, and the questions a personal agent leaves open for a team.

What OpenAI announced

OpenAI introduced dots at DevDay 2026 as "remarkably capable, always-on agents built to handle everything." The design is persistent delegation: you give a dot a goal, connect the apps it needs, set what it may do on its own, and it keeps making progress between conversations and brings completed work back for approval, learning your preferences and standards from feedback over time.

Availability, as of October 2026, per OpenAI's announcement and help center: dots are rolling out to ChatGPT Pro and Business Premium subscribers, with the first dot included in those plans at no extra cost. The Pro rollout excludes the European Economic Area, Switzerland, and the UK for now; Business Premium gets dots in all supported ChatGPT regions. Multiple dots per user, and eventually teams of dots, are planned but not yet available. You create your first dot on desktop (the ChatGPT desktop app or a desktop browser), then message it from the ChatGPT mobile app or connect Slack and Microsoft Teams as messaging channels; texting is coming soon.

What "an agent with its own cloud computer" means

The phrase is literal. Every dot gets a dedicated cloud computer with a browser, separate from your own machine. Three properties follow from that, each verified on OpenAI's documentation as of October 2026:

• Persistence. The computer keeps running when yours does not. Files, installed software, and browser sessions persist on the dot's machine between tasks, so the agent resumes with its working state intact rather than rebuilding it per task.
• Separation by default. Your computer and its contents stay separate unless you explicitly connect them. Access to your local machine is optional and starts turned off, and you can connect only one personal computer at a time.
• Inspectability, for you. You can open your dot's computer at any time to watch its ongoing work, take over control manually, and hand control back. Website logins go through protected sign-in forms, so credentials never reach the model.

This is the same shift the rest of the agent market made through 2026: the agent's execution environment moves off your laptop and becomes a persistent machine in someone's cloud. For the broader taxonomy, see background agents vs cloud agents; for Cloudflare's version of a persistent agent workspace, see Cloudflare OS agent workspaces, explained.

How a dot does coding work

A dot does not replace Codex; it creates and steers Codex tasks. When you ask your dot for code, it files the work as a Codex task, monitors the result, and follows up: in OpenAI's own example flow, you tell your dot to find the cause of a bug, fix it, add a test that fails without the fix, and open a draft pull request, and the dot comes back with the link.

Mechanically, as of October 2026, that works in two ways. For cloud coding, you first set up a Codex cloud environment naming the GitHub repositories it may use; your dot then creates tasks inside that environment, each running in its own sandbox preloaded with the repository. For local coding, a dot can create or continue Codex tasks on a computer you have connected. Repository access goes through the GitHub connector, which lets the agent investigate issues and prepare pull requests in the repositories you granted.

Two accounting details matter for anyone budgeting agent usage. Talking to your dot does not count toward your ChatGPT usage limits, but the tasks it starts in Codex or ChatGPT Work count toward those products' limits, so a busy dot spends your Codex allowance. And OpenAI describes "an allowance for deeper work, with extended limits for the first month after launch," without published post-launch numbers. For what Codex tasks look like once created, see Codex background agents and the agents dashboard.

Permissions: what a dot may do on its own

Dots ship with default rules about which actions run autonomously, which need approval, and which you must do yourself. The most sensitive actions, like changing a password or transferring money, always require you to take over personally. Others, like permanently deleting data or installing software, need approval each time, and some recurring actions can be approved in advance. Custom Rules let you set, per supported action, whether the dot acts freely, acts only when explicitly asked, asks first, or hands off entirely; they cannot switch off the core safety requirements. An automatic review layer checks consequential actions against your instructions and OpenAI's safety requirements, and monitoring can pause or stop a dot's work if it detects a problem.

On data: a dot can receive memories and recent conversation context from ChatGPT, and conversations with it can contribute back to ChatGPT memory, but OpenAI states the dot's context does not retain credentials, images, or screenshots.

Who else can see or drive the session?

Here is the open question for engineering teams. A dot is a personal agent: one person's goals, app connections, approval queue, and cloud computer. ChatGPT Space, announced at the same DevDay, gives teams and agents a shared place for files and project context, and business workspaces get admin controls for managing dots. But the working session itself belongs to its owner: a teammate cannot open your dot's computer, watch the diff take shape, or take the keyboard when the agent heads down the wrong path.

That matters because solo-by-default is already the pattern the data shows. A July 2026 LeadDev analysis of 25,264 agent-generated pull requests across 2,361 popular GitHub repositories found that in 79 percent of agentic PRs the same developer both reviewed and modified the agent's contribution, and only about one in eight workflows involved multiple humans. Personal always-on agents extend what one person can delegate; they do not, by themselves, make agent work visible to the team that has to review, debug, and live with it.

Neither column is wrong; they answer different questions. A dot asks how much of one person's work an agent can carry. A team workspace asks how a group ships software with agents it can watch, steer, and review together.

Where AQ fits

AQ is the multiplayer coding harness where engineering teams run AI coding agents like Claude Code and Codex together: shared live terminals, a code editor, and app previews, in your own cloud. In this guide's terms, AQ is what the always-on cloud computer looks like when it is built for a team instead of a person. Agents run as the real CLIs (Claude Code, Codex, Cursor Agent, Kimi, Grok, or plain shells) in persistent tmux sessions on your team's VM, streamed live to the browser: sessions survive a closed laptop, resume from any device, and teammates open the same workspace and watch the same live session, with typing into someone else's terminal delegated by its owner approving a control request in one click. Each workspace gets its own isolated git worktree with automatic dependency install, agents commit, push, and open PRs with per-user GitHub auth, and every workspace serves a live dev-server preview with shareable links. Visibility is owner-managed: a workspace is team-visible, or private and shared with specific people. Engineers sign into the CLIs with their own Claude and OpenAI accounts (AQ never marks up usage on your own subscriptions), and execution is either VMs you connect from your own cloud or a dedicated always-on AQ-managed VM in its own isolated network, with no shared multi-tenant execution tier. The Free plan is a personal sandbox for one person on a private machine AQ creates, with no time limit; the Team plan is $50 per user per month in early access (standard $200, billed monthly), rate locked for your first 12 months.

Plainly: a dot is a personal chief of staff that happens to file Codex tasks; a coding harness is where the engineering team does the coding work in the open. If your team is already running Codex in the cloud, running Codex on a cloud VM and keeping Codex running after closing the laptop cover the workspace side of the same shift.