OpenAI launches dots, always-on agents powered by GPT-6 Astra

You can reach your dot in ChatGPT on desktop, web and mobile, or message it in Slack and Microsoft Teams. You can jump on a voice call with it too, and OpenAI says texting is coming soon. A dot remembers what's going on across all of those places. So a project you started in ChatGPT can pick up in a Slack thread without you explaining it all again. It can also run several projects at once, so you're not juggling separate threads.

OpenAI's examples are everyday work. A developer's dot keeps an eye on customer feedback, works out small fixes, builds and tests them, and comes back with pull requests and videos showing the changes. A scientist's dot reruns the numbers when new data comes in and updates the figures for a paper. A content creator's dot picks the best clip moments from an interview transcript, writes show notes and lines up social posts for approval.

Inside OpenAI, the company says, dots start digging the moment a bug shows up in Slack. Outside it, one early tester's dot noticed he'd forgotten to invoice a publication. It got the invoice ready and sent it once he said OK.

The safety rails

Safety was the big question hanging over all this. Axios framed the launch as a test of OpenAI's safety promises, after reports of its agents getting around the safeguards meant to hold them back. In August, OpenAI itself reported that AI agents in its July tests broke into parts of Hugging Face.

OpenAI's answer is layers of protection. Each dot works on its own cloud computer, kept apart from yours unless you choose to link them. When you're not actively working with it, it does what OpenAI calls "proactive research", using your connected apps through tools that can only read. OpenAI says that means they can't send messages, change what's in your apps, or take over your browser or computer. Dots can also sign in to supported websites with saved passwords, and the model never sees those passwords.

Dots come with built-in rules for when to act alone and when to check with you first. With Custom Rules, you can allow certain actions, make them need your OK, or ban them outright. An Activity View shows what a dot's up to, background work included. Anything that could affect your accounts or share your information goes through auto-review. That checks the action against your instructions, your rules and OpenAI's safety requirements. "Certain sensitive tasks, such as changing a password, always stay with you," OpenAI says. If its monitoring spots a safety problem, it can pause or stop the dot.

Alexander Embiricos, who's on OpenAI's product staff, told Axios a dot can write a message but shouldn't send it to another person or agent unless the user's asked it to. Axios also reported that dots won't finish some big financial transactions, and hand control back to the user instead. Reuters reported that CEO Sam Altman said dots had been "safety tested".

OpenAI's own advice is simple: "Dots can still make mistakes, so always review consequential work."

On privacy, OpenAI says it doesn't use content from Business, Enterprise or Edu workspaces to improve its models by default. People on personal plans can choose whether their dot's work gets used. OpenAI says it doesn't train directly on proactive research or on the notes a dot keeps for itself.

What it costs and who gets it

The first dot comes with Pro and Business Premium. Chatting with it doesn't count against your ChatGPT usage limits, but tasks it kicks off in Codex or ChatGPT Work do. OpenAI says plans get higher limits for the first month after launch. Later on, it says, you'll be able to add more dots, or pay to make a dot faster or let it take on more work each month.

To start, you set up a dot in the ChatGPT desktop app or a desktop browser, connect your apps and let it introduce itself. After that you can use it in the mobile app as well.

OpenAI also gave a first look at specialist dots. A company sets these up with their own identity, logins and system access to handle set jobs. OpenAI's starting with small enterprise pilots, built on its own testing in procurement, invoice processing, email marketing, customer support and commercial contracts. It's also working with Microsoft to bring specialist dots under the governance and security controls in Microsoft's Agent 365.

The rest of DevDay

Dots topped a long list. OpenAI's own DevDay recap counts more than 20 major announcements across ChatGPT, Codex and its models. OpenAI launched GPT-6.1 Sol, which it says comes close to Astra on coding, using a computer and professional tasks, at one-fifth of Astra's standard token prices. OpenAI's API documentation lists it at US$2 per million input tokens and US$10 per million output tokens.

OpenAI also brought in a Pro 500 plan with 25 times the ChatGPT Plus usage allowance, an Ultrafast speed tier for Astra, computer use in its Agents API, and ChatGPT Space. That's a shared workspace where teammates and their dots work from the same documents. And it repeated its claim that it hit its September goal of an automated research intern, which we covered when the company first made it.

Reuters framed dots as OpenAI's answer to Meta's Muse agent, in the race to sell AI that can work with little supervision. For now, each Pro or Business Premium user gets one dot. OpenAI says it plans to bring dots to more people soon, and that down the track it sees teams of dots working together. You can watch the full keynote and a short official demo, "Introducing dots, always-on agents built to handle everything", on OpenAI's YouTube channel.