OpenAI Dots: Always-On Agents With Their Own Computers

OpenAI just made the chatbot a coworker that never logs off. At DevDay on September 29, 2026, the company launched "dots": always-on agents inside ChatGPT that each get their own cloud computer and web browser, run on GPT-6 Astra, and keep working toward your goals after you close the laptop. The real story is not the feature list. It is that the unit of AI product is shifting from a conversation to a worker you leave running.

• A dot is a persistent agent on GPT-6 Astra with a dedicated cloud computer and browser, running 24/7 against goals you set.
• It connects to 4,000+ apps through plugins and keeps context across ChatGPT, Slack, Microsoft Teams and voice calls.
• Rollout starts with ChatGPT Pro ($200) and Business Premium, one dot per user, first dot included at no extra cost.
• Approval gates stay on: a dot cannot send messages, edit content or control computers without your sign-off.

What did OpenAI actually launch at DevDay?

Dots were the headline of a 20-plus announcement day. Each dot is an agent that lives on GPT-6 Astra, OpenAI's flagship, and owns a dedicated cloud computer and web browser. You give it a goal, it works toward it around the clock, and it does not need your device to stay awake. When it has nothing assigned, it does what OpenAI calls proactive research, using read-only tools, and it learns your preferences over time.

RelatedGPT-6.1 Sol: OpenAI's Cheap Model Nears Astra's Power

The rest of the day fills in the workspace around it. ChatGPT Space is a shared workspace where teammates and their dots operate together. Pages is a document type that humans and dots edit jointly, covering images, writing and charts. Codex now runs in the cloud from any device, phones included, and its CLI accepts voice. OpenAI also previewed OpenAI Private Intelligence, which promises stronger data controls, and launched GPT-6.1 Sol the same day. We already broke that one down in our GPT-6.1 Sol coverage.

How does a dot work day to day?

Think of it as a contractor with its own desk. The dedicated computer and browser matter because the agent no longer borrows your session or your machine. It can sit in a tab, use the tools you have connected, and keep going while you sleep. Plugins reach more than 4,000 apps, and the same dot keeps context whether you ping it in ChatGPT, Slack, Microsoft Teams or over a voice call. That continuity is the quiet feature. Most agents forget who you are between surfaces, and this one is built not to.

Control is handled with Custom Rules. You define what the dot may do alone, when it must ask first, and what it must never do. Sending messages, editing content and controlling computers all require approval, and sensitive tasks like changing passwords are left for you to handle manually.

The blanks in that table are the point. Meta launched its Muse assistant on September 8 at $20 to $100 a month, and neither company has published anything comparable on always-on execution in the material we reviewed. OpenAI is defining the category before anyone else has a listed answer.

Who can get dots, and where?

Rollout starts with ChatGPT Pro at $200 and Business Premium, in eligible markets. The first dot is included at no extra cost, and chatting with it does not count against your ChatGPT usage limits. Pro users in the EEA, Switzerland and the UK are excluded at launch. Enterprise, Edu and Healthcare workspaces get a beta once an admin switches it on. Everyone is capped at one dot per user for now, with multiple agents and scaling planned for later.

Why does the Pro multiplier cut matter?

Tucked into the same announcements is a plan change. The Pro $200 tier's usage multiplier drops from 20x to 10x Plus for new subscribers, while existing ones keep 20x. The Pro $100 tier stays at 5x, and sign-ups reopen after the September 10 pause. Read that next to the dots news and the logic is clear: an agent that runs all day burns compute that chat never did. Exempting dot conversations from usage limits while trimming the headline multiplier shifts the value from more chat to more autonomy. If you bought Pro for raw message volume, the new deal is worse. If you want a worker, it is better.

The same logic explains the Ultrafast tier. It runs up to 8x faster in Codex and up to 6x faster in the API at 6x the standard API price, available for GPT-6 Astra on Pro 500 and Enterprise, with a GPT-6.1 Sol Ultrafast at 350 tokens per second coming soon. Speed is now a priced product. To see how models stack up on code, check our AI coding leaderboard.

RelatedChatGPT Gets Sponsored AI Agents, Angi Is the First Advertiser

What can go wrong with an always-on agent that has your apps?

Plenty, and the approval model is the whole defense. An agent holding connections to thousands of apps and a real browser is a bigger target than a chat window. The mitigation OpenAI describes is sensible: no sending, editing or controlling without sign-off, plus rules you write yourself. But approval fatigue is real. If a dot asks permission forty times a day, people click yes on autopilot, and the gate stops being a gate. Custom Rules only help if users take the time to set them well.

The timing colors it too. Days earlier OpenAI halted the release of GPT-6.1 Astra over safety concerns, which we covered in that report. Shipping a persistent agent on the current flagship days later is a bet that the guardrails, not the model, are what matter.

What does it mean for the market, and who is exposed?

The signal for investors is that agents with their own compute become a recurring-revenue surface, not a chat feature. Microsoft (MSFT) is OpenAI's partner, and dots that live inside Teams tie its collaboration stack directly to OpenAI's agent layer, though how the economics split is not disclosed. Alphabet is exposed on two fronts: the workspace tools a dot can operate on your behalf, and a browser-driving agent model that competes with its own products. Meta, with Muse at $20 to $100 a month, is the nearest consumer rival and has no announced equivalent. None of this is a recommendation to buy or sell anything. It is a reminder that the contest is moving from who has the best model to who owns the always-running worker.

Our take

Dots are the right idea shipped carefully, and the careful part is what keeps me interested. A chatbot that answers is a tool. An agent with its own desk is closer to staff, and staff need a manager. The approval gate makes you that manager, which is both the safety feature and the friction. I think the product wins or loses on the quality of its defaults, because most people will never write a thorough rule set. Watch whether OpenAI ships sensible presets. If it does, a dot in every Pro account is a plausible default by next year. If it does not, expect a headline about a dot doing something nobody approved.

Original analysis by GenZTech, based on OpenAI and DevDay 2026 coverage.