OpenAI ships Dots, always-on agents built on GPT-6 Astra

OpenAI ships Dots, always-on agents built on GPT-6 Astra

OpenAI launched Dots on 29 September 2026 at its DevDay event. Dots run on GPT-6 Astra, get their own cloud computer and browser, and connect to more than 4,000 apps through OpenAI's plugin ecosystem. The company is rolling dots out today to ChatGPT Pro and Business Premium users in eligible markets, with enterprise pilots for "specialist dots" built for defined jobs such as invoicing and customer support.

Dots are OpenAI's always-on AI agents: persistent assistants that run on GPT-6 Astra, keep a dedicated cloud computer and browser, and continue tasks in the background between conversations. They connect to a person's apps through Slack, Teams and ChatGPT, learn preferences over time, and return finished work for review rather than waiting to be asked.

What changed

A dot reaches a person through ChatGPT on desktop, web and mobile, through Slack and Microsoft Teams, or through a voice call, and OpenAI CEO Sam Altman said at the DevDay keynote that phone and text access are coming soon, according to VentureBeat.

• The first dot is included at no extra cost on Pro and Business Premium plans, with an extended usage allowance for the first month after launch, per OpenAI's announcement.
• Between conversations, a dot runs "proactive research": read only tools that gather information from connected apps but cannot send messages, change app content, or control a browser or computer.
• OpenAI is previewing "specialist dots" with their own identity, credentials and IT provisioned hardware for tasks such as procurement, invoice processing, email marketing, customer support and commercial contracting.
• OpenAI is working with Microsoft to bring specialist dots into Microsoft Agent 365's enterprise governance and security controls.
• Dots ship alongside ChatGPT Space, a shared workspace where employees, ChatGPT, Codex and dots work from the same project files and pages, replacing the existing ChatGPT Library for Pro, Business and Enterprise users, according to VentureBeat.
• Enterprise, Edu and Healthcare workspace admins can enable a beta for their users.

Why it matters for agent builders

Dots publish a permission model that agent builders can study or copy. A system called Auto-review checks every planned action, such as sending an email or changing a file, against the user's instructions, their Custom Rules, and OpenAI's safety requirements before it runs, and it tells the dot why a step was blocked so the dot can try an approved alternative or hand the step back, according to OpenAI's safety writeup. Some actions, including permanently deleting data, installing software from an unrecognized source, changing a password or moving money between accounts, always require the person, not the dot, to do them.

Credentials get a separate safeguard. For supported sign ins, the model pauses while a person enters a password directly into the browser environment, so the password never enters the model's context, and a dedicated encrypted credential service supplies saved passwords the same way, per the same writeup. Each dot's cloud workspace is sandboxed and isolated from the systems that enforce those checks, so a dot cannot use its own access to turn off the review that constrains it.

TechCrunch's Lucas Ropek notes that much of what dots do was already possible through Codex and similar agentic harnesses, and that the product's contribution is bundling continuity, memory and multi app access into one persistent identity rather than a single new capability. For anyone building an agent that needs to act across a person's tools without a human re-prompting it every time, that bundle, not any one feature in it, is the part worth studying.

What to watch

OpenAI has not published pricing beyond the first dot included in Pro and Business Premium, saying only that people will later be able to add more dots and scale each one's speed or monthly workload. The Microsoft Agent 365 integration for specialist dots is described as a work in progress rather than a shipped feature. Given open reports of AI agents crossing access boundaries in other products this year, whether Auto-review and the read only limits on proactive research hold up against real prompt injection attempts, once dots run unsupervised at scale, is the open question for anyone deciding how much to delegate to one.

Key takeaways

• OpenAI launched Dots on 29 September 2026, always-on agents running on GPT-6 Astra with their own cloud computer, browser and connections to over 4,000 apps.
• Dots are rolling out today to ChatGPT Pro and Business Premium users in eligible markets, with the first dot included at no extra cost.
• Auto-review checks planned actions against a user's instructions, Custom Rules and OpenAI's safety requirements before they run, and some actions, like password changes or money transfers, always go back to the person.
• Specialist dots, previewed for enterprise pilots, get their own identity and credentials, and OpenAI is building Microsoft Agent 365 integration for them.
• Dots ship with ChatGPT Space, a shared workspace where people, ChatGPT, Codex and dots work from the same files and pages.

FAQ

What model powers OpenAI's dots?

Dots run on GPT-6 Astra, the model OpenAI describes as its most capable and most aligned, trained to follow a user's intent and refuse harmful requests including biological or cybersecurity misuse, according to OpenAI.

Which ChatGPT plans get dots first?

Dots are rolling out from 29 September 2026 to Pro and Business Premium users in eligible markets, with the first dot included in the plan at no extra cost. Enterprise, Edu and Healthcare workspaces can try a beta once their admin enables it, per OpenAI's announcement.

Can a dot change a password or move money on its own?

No. OpenAI's action rules keep certain sensitive steps, including changing a password and transferring money between financial accounts, with the person rather than the dot, and Custom Rules cannot override that requirement, according to OpenAI's safety writeup.

Source: OpenAI Discussion: Hacker News

Builders comparing always-on agent products and the MCP servers that connect them to other tools can browse the current field on AI Agents Listing.