- OpenAI Dots is an always-on AI agent for ChatGPT that keeps working after you log off
- Each Dot runs on GPT-6 Astra with its own cloud computer and browser
- Access rolls out to Pro and Business Premium plans first, with a workspace opt-in for Enterprise, Education, and Healthcare
- You get one Dot per eligible plan for now; multi-Dot teamwork is on the roadmap
- Granular permissions, activity view, and approval rules keep your Dot under control
What Is OpenAI Dots? Always-On AI Agents Explained
OpenAI Dots explained in one sentence: Dots are persistent AI agents inside the ChatGPT ecosystem that accept projects, keep working in the background, and check in with you only when they need input. Where a standard chat session ends the moment you close the tab, a Dot keeps running on its own cloud infrastructure — even when your laptop is asleep, locked, or powered down.
The headline idea is autonomy without babysitting. You are not expected to monitor or repeatedly correct a Dot. Instead, you assign work, walk away, and review results when the Dot reports back.
Video Highlights:
- Dots keep working after you leave — the core "always on" pitch
- One person's Dot and a teammate's Dot can collaborate on a shared goal
- Powered by GPT-6 Astra with a dedicated cloud computer and browser
- Over 4,000 plugins extend what a Dot can touch
- Dots learn your preferences, goals, and definition of a good result
Think of a Dot less like a chatbot and more like a remote assistant with its own computer: you delegate, it executes, and it pings you when a decision is required.
Core Capabilities at a Glance
OpenAI is positioning Dots around a handful of pillars: background execution, multi-project handling, plugin access, preference learning, and two-way communication. The comparison table below summarizes how each capability maps to real-world use.
| Capability | What It Does | Practical Example |
|---|---|---|
| Always-on execution | Keeps working after you disconnect | Long research or drafting tasks run overnight |
| Multi-project handling | Several projects per Dot, no separate chats | One Dot juggles invoices, research, and scheduling |
| Dot-to-Dot teamwork | Your Dot collaborates with teammates' Dots | Two Dots split a shared deliverable |
| 4,000+ plugins | Connects to external tools and services | Pulls data from apps, files, and services |
| Local file access | Works with files on your computer | Edits or reads documents you authorize |
| Preference learning | Remembers goals and quality standards | Formats outputs the way you like over time |
Background Worker
- Runs on its own cloud computer
- Survives laptop sleep and shutdown
- Check progress anytime, or don't
Proactive Assistant
- Notices forgotten tasks, like an unsent invoice
- Prepares work and asks for approval
- Contacts you with questions, not just replies
Personalized Agent
- Learns from your feedback over time
- Remembers goals and preferences
- Adapts to your definition of a good result
Communication flows both directions. You can message or call your Dot inside ChatGPT, and the Dot can reach out to share progress, ask a question, or request a decision when it hits a fork in the road.
Security and Control: What You Can Manage
Because a Dot is always on, OpenAI is emphasizing transparency and permission controls. This matters most for small and medium businesses without large IT departments monitoring every automated action.
| Control Layer | Function | When It Matters |
|---|---|---|
| App access rules | Allow, require approval, or block per app | Limiting a Dot to sanctioned tools |
| Activity view | See exactly what your Dot is doing | Catching unauthorized access attempts |
| Approval system | Sensitive or risky actions need your sign-off | Preventing compromised-account damage |
| Admin opt-in | Workspace admins enable Dots for orgs | Enterprise, Education, Healthcare rollouts |
When you first get access, spend time in the activity view before delegating sensitive work. It is the fastest way to learn what your Dot attempts on its own — and to tighten app rules before problems arise.
Audit App Permissions
Go through the allow / require-approval / block rules and restrict your Dot to the apps it genuinely needs.
Watch the Activity View
Check what actions your Dot takes in its first days, especially any access attempts you did not authorize.
Set Approval Thresholds
Configure which actions — payments, emails, data sharing — must wait for your explicit approval.
Delegate Gradually
Start with low-stakes projects and expand scope as the Dot's behavior matches your expectations.
Access and Plan Eligibility
Availability is rolling out in waves. Here is how eligibility breaks down at launch, based on OpenAI's rollout statements.
| Plan Tier | Dots Access | Notes |
|---|---|---|
| Pro | Rolling out | First Dot included with allowance for deeper work |
| Business Premium | Rolling out | First Dot included with allowance |
| Enterprise | Beta available | Requires workspace admin to enable |
| Education | Beta available | Requires workspace admin to enable |
| Healthcare | Beta available | Requires workspace admin to enable |
| Free / Plus | Not yet announced | No confirmed timeline at time of writing |
Even on eligible paid plans, you can currently run only a single Dot. Multiple Dots working together is a stated future direction, not a shipping feature, so plan your delegation around one agent.
How Dots Compares to Rival Agents
Dots land in an increasingly crowded agent landscape. The table below positions them against the most visible competitors discussed around the launch.
| Agent | Maker | Key Differentiator | Availability Note |
|---|---|---|---|
| Dots | OpenAI | Always-on cloud execution, 4,000+ plugins | Pro and Business Premium rollout |
| Grok Bots | xAI | Agent features inside X ecosystem | Regional availability varies |
| Muse | Meta | Consumer agent assistant | Not available in all regions (e.g., UK) |
Treat any comparison as provisional. Dots' live demos — including the message-and-call feature — did not always work smoothly during the presentation, so expect rough edges during the rollout period.
Your First Week with a Dot: Setup Checklist
Once Dots appear on your account, a structured first week builds confidence in the agent. Work through this checklist to get value quickly without overexposing your accounts.
First Week Goals:
- Verify Dots access appears on your plan
- Lock down app permissions before delegating
- Assign one low-risk background project
- Review the activity view daily for a week
- Give explicit feedback so the Dot learns your preferences
Dots are designed to learn from your corrections, goals, and quality standards. The more feedback you give early, the faster your Dot converges on outputs that match your expectations.
FAQ
Q: What is OpenAI Dots in simple terms?
OpenAI Dots are always-on AI agents for ChatGPT. You give a Dot a project, and it keeps working in the background on its own cloud computer and browser — even when your laptop is asleep or off — then reports back with progress or questions.
Q: Which plans get access to Dots?
Dots are rolling out to Pro and Business Premium users, with the first Dot included alongside an allowance for deeper work. Enterprise, Education, and Healthcare users can try a beta if their workspace administrator enables it.
Q: Can I run multiple Dots at once?
Not yet. For now you can have only one Dot per account. OpenAI has said multiple Dots working together is a future direction, and demos showed one person's Dot collaborating with a teammate's Dot, but that is not a shipping feature today.
Q: What powers a Dot, and what can it access?
Dots run on GPT-6 Astra and come with their own cloud computer and browser. They support over 4,000 plugins, can work with local files on your computer, and you control access per app with allow, approval-required, or block rules.
Q: How do I keep a Dot from doing something risky?
Use the app access rules to restrict what your Dot can touch, monitor the activity view to see its actions in real time, and rely on OpenAI's review system, which flags sensitive actions — such as anything that could compromise your account — for your approval.