Dots, OpenAI's Always-On Agents, Bring a Dedicated Cloud Computer to Every User
OpenAI has introduced Dots, a class of always-on AI agents that run on their own cloud computer, connect to thousands of apps, and keep working toward your goals whether or not you're looking at a screen.
OpenAI has introduced Dots, a class of always-on AI agents that run on their own cloud computer, connect to thousands of apps, and keep working toward your goals whether or not you're looking at a screen.
A computer that never logs off
The core change with Dots is structural rather than cosmetic. Each dot is assigned its own cloud computer, separate from the session you happen to have open in ChatGPT, Slack, or Teams. That separation is what allows a dot to keep working on a task after you close your laptop, and to work on multiple projects concurrently without one task blocking another. The agent is powered by GPT-6 Astra, and it can use its own browser and any connected apps to actually carry out tasks, not just draft suggestions about them. Because the computer persists independently of any single app window, the same dot can pick up a task in Slack and continue it later in ChatGPT or Teams, carrying its context along rather than starting over.
Reach across thousands of apps, with limits you set
Dots can connect to more than 4,000 apps through plugins, which is what turns an agent from a chat window into something that can actually move data and take action across the tools an organization already uses. OpenAI has paired that reach with explicit control mechanisms. Users can define permissions and approval requirements through Custom Rules, deciding in advance which actions a dot may take on its own and which require sign-off. Proactive research, where a dot goes looking for information without being asked in the moment, is restricted to read-only mode on connected apps, so a dot doing background research cannot also make changes while it looks around. Sensitive tasks, such as changing a password, are explicitly kept with the user rather than handed to the agent.
Watching the work, not just the output
One of the more concrete design choices is that users can inspect a dot's computer and see its work directly, rather than only receiving a final summary. That visibility matters alongside a built-in safety behavior: a dot can pause or stop work when its own monitoring detects a safety concern, rather than continuing on a course that looks risky. Dots also learn user preferences through feedback over time, which means the system is meant to adjust its behavior based on how a person responds to what it has done, not just execute a fixed script indefinitely.
Specialist dots and where the identity question goes next
Beyond a single general-purpose agent, OpenAI is building specialist dots that can have their own separate identities, credentials, and organizational responsibilities, distinct from the person who set them up. These specialist dots are being integrated with Microsoft Agent 365, which points toward a future where an agent's identity and permissions are managed inside the same infrastructure that already governs human accounts inside a company. That integration is still early, but it signals that the questions OpenAI is trying to answer with Custom Rules and read-only research modes extend into how organizations will eventually manage many agents with distinct roles, not just one assistant per person.
Who gets a dot, and what it costs to find out
Access follows a fairly specific set of conditions. Dots are rolling out to Pro and Business Premium users in eligible markets, with the first dot included in those plans at no extra cost. Enterprise, Edu, and Healthcare users can reach a beta version, but only once a workspace administrator enables it, which puts the initial decision in the hands of IT or account administrators rather than individual users. On the usage side, conversations with a dot do not count toward ordinary ChatGPT usage limits, but tasks that get started or managed through Codex or ChatGPT Work still count toward normal limits. For anyone already on a Pro or Business Premium plan, that means the practical next step is simply checking whether a dot has appeared in their ChatGPT, Slack, or Teams interface, and for enterprise users, asking their administrator whether the beta has been switched on.
Specs
- Model: GPT-6 Astra
- App integrations: Over 4,000 apps
- Availability: 24/7
- Included plans: Pro and Business Premium
Sources
- OpenAI — OpenAI