OpenAI Dots: Always-On AI Agents With Their Own Cloud Computer

OpenAI launched dots at DevDay 2026: persistent, always-on agents that keep working between conversations, run on their own cloud computer with their own browser, and can connect to more than 4,000 apps through the plugin ecosystem. Powered by GPT-6 Astra, a dot is pitched as an agent that learns your preferences over time, takes on multiple projects in parallel, and sometimes brings you finished work before you ask — OpenAI cites an early tester whose dot noticed an un-invoiced publication, prepared the invoice, and sent it after his approval.

This piece covers how dots actually work, the safety boundaries OpenAI built around them, who gets them today and at what cost, and where the specialist enterprise variant is headed. It is written for engineers and team leads deciding whether to wire real work into an always-on agent.

Dots: what OpenAI launched

A dot is not a chat session. It persists across conversations, works toward goals 24/7, and is reachable wherever you work: ChatGPT on desktop, web, and mobile, plus Slack and Microsoft Teams, with texting coming later. Context carries across every channel, so a project started in ChatGPT can be steered from a Slack thread. You can name your dot, give it feedback, and hop on a voice call with it when typing is too slow.

Inside OpenAI, the company says dots investigate bugs the moment they appear in Slack, turn new designs into working apps while the team handles customer feedback, and keep planning cycles in sync. The launch model is one primary dot per user now, with teams of dots envisioned later — and a preview of specialist dots that take on defined organizational roles like procurement, invoice processing, and customer support.

How a dot works: its own computer, browser and apps

Each dot runs on its own cloud computer, separate from yours. You can open that computer at any time to inspect what the dot is doing, and optionally grant it permission to connect to your own laptop and work alongside you. For websites that need authentication, dots can sign in with saved passwords without the password ever being exposed to the model itself.

When you are not actively working with it, a dot performs what OpenAI calls proactive research: scanning the apps you have already connected for ways to be useful. Those background tools are deliberately restricted to read-only access — a dot doing proactive research cannot send messages, change app content, or control a browser or computer. Real actions happen only inside active work, under the permission system.

Boundaries and safeguards: what a dot cannot do

Every consequential action a dot takes passes through auto-review, which checks it against your instructions, your Custom Rules, and OpenAI's safety requirements to decide what proceeds automatically, what needs your approval, and what stays with you entirely — changing a password, for example, always remains a human task. Custom Rules let you allow specific actions outright, require approval, or block them, and Activity View shows everything the dot is doing, including background work.

OpenAI layered monitoring on top: security safeguards watch for malicious instructions and potentially harmful behavior, and the monitoring system can pause or stop a dot's work if it detects a safety concern. On the data side, Business, Enterprise, and Edu workspace content is not used for training by default; personal plans get a control; and OpenAI says it does not train directly on proactive research or a dot's notes to itself. The company still warns that dots make mistakes and consequential work needs review.

Availability and pricing: who gets a dot today

Dots are rolling out now to Pro and Business Premium subscribers in eligible markets. Enterprise, Edu, and Healthcare workspaces can try the beta once an admin enables it — it ships off by default. The first dot is included in a Pro or Business Premium plan at no extra cost, with an allowance for deeper work and extended limits during the first month after launch.

The usage accounting matters for anyone budgeting: conversations with your dot do not count toward ChatGPT usage limits, but Codex or ChatGPT Work tasks the dot starts count as usual. Setup is desktop-first — you create your first dot in the ChatGPT desktop app or a desktop browser, connect apps, and then message it from mobile afterward. In the future OpenAI plans to let you add more dots and scale each one by speed or by total monthly work.

Specialist dots and what to watch next

For organizations, OpenAI previewed specialist dots: agents with their own identity and credentials for access management, IT-provisioned hardware, and deep integrations with systems of record, taking on well-defined responsibilities. Rollout starts with focused enterprise pilots in which OpenAI's engineering teams work directly with customers to define each dot's duties, tools, and review process — informed by internal testing across procurement, invoicing, email marketing, support, and contracting. A partnership with Microsoft will bring specialist dots under Agent 365's governance and security controls.

The open questions are the ones that decide whether dots become infrastructure or novelty: how "eligible markets" expands, what adding a second dot costs, how the deeper-work allowance behaves after the first month, and whether the read-only proactive research boundary holds up in practice. Anyone planning to delegate real responsibilities should start with low-stakes work, keep approval requirements on, and watch the Activity View closely for the first few weeks.