OpenAI Dots brings always-on agents with dedicated cloud computers to ChatGPT

Persistent agents, not longer chats

On 29 September 2026 OpenAI used its DevDay event to announce Dots, a new kind of agent inside ChatGPT that keeps working on assigned goals between conversations. According to a setup guide published on dev.to, a Dot is not merely a chat session that persists: each one gets its own cloud computer, browser and memory, runs on a model OpenAI calls GPT-6 Astra, and connects to more than 4,000 apps through plugins.

Instead of requesting one-off analysis, you hand a Dot ongoing responsibility for a job — for instance keeping a sales proposal current as requirements shift, investigating a bug as soon as it appears, or preparing content when new source material arrives. The agent monitors connected apps in the background, reports progress, asks questions when unsure, and returns decisions that need human judgement. You can open its cloud computer at any time to see what it is doing, and the desktop app adds an Activity View showing the steps it has delegated.

Dots are reachable through ChatGPT on desktop and mobile, via Slack and Microsoft Teams, or by voice call. Texting support currently exists only as a small US Pro beta.

Availability and pricing

Per dev.to, access at launch is limited to paid tiers. ChatGPT Pro plans at $100, $200 and $500 per month include a first Dot at no extra cost, but only for adult accounts, and not in the EEA, Switzerland or the UK. ChatGPT Business Premium — $100 per user per month billed annually, or $125 monthly — gets Dots in all supported regions. Enterprise, Edu and Healthcare workspaces have Dots in beta but disabled by default until a workspace admin enables them. Free, Go ($8) and Plus ($20) plans are excluded; OpenAI says it plans to expand access later without giving a date.

Talking to a Dot does not count against ChatGPT message limits, but work it delegates to ChatGPT Work or Codex counts as usual. Pricing for additional Dots has not been published, and rollout is gradual — the dev.to guide notes an eligible account may wait several days before Dots appear.

Setup and autonomy rules

Creation is desktop-only, through the ChatGPT desktop app for macOS or Windows or via chatgpt.com/dots in a desktop browser; once a Dot exists, it can be messaged from the mobile app. The name you give it becomes its handle (@yourname-alfred), which is how you mention it in Slack, Teams and ChatGPT.

Apps are linked in a Plugins tab, and permissions are shared across ChatGPT, Work and Codex, so each connection deserves review before approval. Custom Rules define how much autonomy each Dot has, with four levels described by dev.to: take action on its own, act only on pre-approved items, ask first every time, or hand the step back to a human. Access to your local computer exists but is switched off by default.

Guardrails and limits

A Dot's background research is read-only by default — it cannot send messages or change data while simply watching your apps. OpenAI also hard-codes two constraints: password changes and money movement always stay with a human, and a separate Auto-review system examines planned actions before they run. On prompt injection, OpenAI reports no scored successes across 16,600 generated attack emails, though the guide cautions that scope problems grow as task chains get longer. Its practical advice is to start conservative, keeping "ask first" on anything consequential and reviewing important work early on.

Why it matters

Dots shift ChatGPT from a reactive chatbot toward delegated, always-on software that owns a job rather than an answer. Giving each agent its own machine turns the abstraction from an AI feature into something closer to a remote worker with tools, credentials and standing instructions. That raises the stakes on permissions — shared across ChatGPT, Work and Codex — and helps explain OpenAI's caution: regional exclusions, adults-only tiers, admin opt-in for large organisations, and hard-coded human checkpoints for money and passwords. For developers, a persistent agent with plugin access to thousands of apps is also a signal to build integrations that assume background, autonomous access rather than a single conversation at a time.