OpenAI is rolling out dots, a new class of always-on AI agents that take on work autonomously, starting today for users on its Pro, Business Premium, and Enterprise plans. Each dot runs on its own cloud computer and is powered by the company's GPT-6 Astra model. The rollout is limited to eligible markets at first, with broader availability promised soon.
The agents are reachable through ChatGPT on desktop, web, and mobile, and also through Slack and Teams. Texting access is coming later. Users can message a dot or speak with it by voice call, and it keeps context across every channel. Plugins let a dot connect to more than 4,000 apps, and it can use its own browser and connected apps to get work done.
Read more: OpenAI Launches Its First Hardware Product, a $230 Macro Pad for Coding Agents
Every user starts with one primary dot and can give it a name. OpenAI says teams of dots are envisioned over time. A dot can run one project while juggling several others, and it works toward a user's goals around the clock. When the user is not actively working, proactive research runs in the background, restricted to read-only tools that cannot send messages or change app content.
Users can open a dot's computer at any time to inspect what it is doing, and an Activity View tracks progress, including background work. A dot can also connect to devices beyond its own computer, and users can grant it permission to use their laptop. When signing in to supported sites, a dot can draw on saved passwords without the model seeing them, and a monitoring system can pause or stop a dot's work.
Permissions are managed through the existing ChatGPT app controls, with Custom Rules that allow, require approval for, or block specific actions. Built-in safety requirements always apply, and an auto-review checks actions that affect accounts or share information. Sensitive tasks such as changing a password always stay with the user. OpenAI cautions that dots can still make mistakes and says consequential work should be reviewed.
For business customers, OpenAI says it does not train on ChatGPT Business, Enterprise, or Edu workspace content by default. Personal plan users control whether dot conversations are used for model improvement, and there is no direct training on proactive research or a dot's notes. An early tester's dot prepared and sent an invoice after receiving approval, according to the announcement.
OpenAI also previewed specialist dots with their own identity for access management, including IT-provisioned hardware and deep integrations with systems of record. These begin with focused enterprise pilots, with OpenAI engineering teams working directly with organizations. OpenAI says it is working with Microsoft so that Agent 365 security and governance controls can cover specialist dots, with the aim of letting companies oversee them using Microsoft tools already in place.
The launch follows OpenAI's recent model releases, including GPT-6.1 Sol, which the company said offers near-Astra intelligence at one-fifth of Astra's standard API token prices. GPT-6 Sol and Luna, announced earlier, were described by the company as two models striking different balances between capability and cost.
