What Is OpenAI Dots? Always-On AI Agents Explained

OpenAI's new "dots" agents run on their own cloud computers and keep working on your goals after you close the app. Here's how they work, what they cost, how they compare to Meta's Muse, and the safety questions raised at launch.

Technology 30 September, 2026 13 min read

1. What Is OpenAI Dots?

At its annual DevDay event in San Francisco on September 29, 2026, OpenAI introduced dots — always-on AI agents built on its newest flagship model, GPT-6 Astra. Unlike a normal ChatGPT conversation, where you send a message and wait for a reply, a dot is assigned an ongoing goal and keeps working on it independently, even while your device is off.

Each dot runs on its own dedicated cloud computer, remembers context across sessions, and can take action inside more than 4,000 connected apps. Rather than asking for approval at every step, a dot checks back in only when it has something to show you or needs a decision it isn't authorized to make on its own.

Always-On

A dot keeps working on its assigned goal continuously, not just during an active chat session.

Own Cloud Computer

Each dot runs on dedicated cloud infrastructure rather than sharing a session with your chat window.

Persistent Memory

Context carries forward between check-ins, so a dot builds on prior progress instead of starting over.

4,000+ App Connections

Dots can read and act inside thousands of third-party apps through OpenAI's plugin ecosystem.

2. How OpenAI Dots Works

The core shift with dots is moving from a request-and-response pattern to a delegate-and-review pattern.

Figure 1: How a dot operates — you assign an ongoing goal, it works continuously on its own cloud computer, and it checks back in only when a decision is needed.

Why This Is Different From a Normal ChatGPT Chat

A standard ChatGPT conversation is synchronous — you ask, it answers, the session ends when you close the tab. A dot is designed for work that unfolds over hours or days: tracking an inbox, maintaining a document, or following up on a multi-step process. It doesn't need you present the whole time, only at the decision points you've defined.

3. Key Features of Dots

Feature	What It Means
Powered by GPT-6 Astra	OpenAI's flagship model for long-horizon, agentic, computer- and browser-use tasks, released September 3, 2026
Dedicated cloud computer	Each dot runs on its own cloud instance rather than sharing your chat session
App connections	Access to more than 4,000 apps through OpenAI's plugin ecosystem
Multi-channel access	Available inside ChatGPT, Slack, and Microsoft Teams, plus voice calls and text (limited beta)
Permission controls	You define what a dot can do autonomously versus what requires your explicit approval
Learns from feedback	Improves at a task over time based on corrections you provide

4. OpenAI Dots Pricing and Availability

Dots Access

1 dot included free with ChatGPT Pro or Business Premium

A new $500/month tier has been introduced for additional dot capacity. Pricing for extra dots beyond the first, and for faster processing speeds, was not fully detailed at launch.

Plan / Access	Dots Included	Notes
ChatGPT Pro	1 dot included	Available outside the EEA, Switzerland, and UK at launch
Business Premium	1 dot included	Same regional availability as Pro
Additional capacity tier	Extra dots / speed	~$500/month; further per-dot pricing unannounced
Enterprise	Beta access	Disabled by default; rolling out gradually
GPT-6 Astra API (developers)	N/A — model access	$10 per million input tokens, $50 per million output tokens

Not Yet Available Everywhere

At launch, dots access was rolled out to ChatGPT Pro and Business Premium subscribers outside the EEA, Switzerland, and UK, and enterprise access remained in a limited, opt-in beta. Availability is expected to expand over time, but check current regional access before assuming your account qualifies.

5. OpenAI Dots vs Meta Muse

Dots arrived as OpenAI's direct answer to Meta's Muse, an always-on personal AI agent that had already climbed to the top of Apple's App Store and Google Play before Dots launched. Both compete on the same basic pitch — an assistant that keeps working after you close the app — but they target different audiences.

Figure 2: OpenAI is positioning Dots for enterprise and business workflows, while Meta is marketing Muse directly to consumers through its existing social apps.

Factor	OpenAI Dots	Meta Muse
Primary audience	Enterprise / business	Consumer
Underlying model	GPT-6 Astra	Meta's in-house agent model
Distribution	ChatGPT Pro / Business Premium	Facebook, Instagram, WhatsApp reach
Isolation model	Dedicated cloud computer per dot	Per-user virtual machine
Permission system	Configurable autonomy rules per dot	"Sentinel" layer approves/blocks/escalates actions
App integrations	4,000+ apps via plugins	Meta's own app ecosystem plus connected services
Early traction	Just launched at DevDay 2026	Already top of App Store and Google Play charts

6. How Dots Fits Among Other AI Agents

Dots enters a fast-moving field of "always-on" or agentic AI products. Beyond Meta's Muse, competitors and adjacent products in this space include Grok (xAI's assistant, with its own bot-style automation features) and Anthropic's Claude, which offers its own agentic and coding-focused tools. OpenAI's own Codex product, focused on software engineering tasks, is a related but separate offering that can also connect into dots-style workflows for coding-specific work.

Context: This Launched Amid a Broader Model Update

Dots was announced shortly after OpenAI shipped its flagship GPT-6 Astra model (released September 3, 2026), which OpenAI describes as built for demanding, end-to-end work including advanced analysis, software engineering, deep research, and long-horizon tasks involving computer and browser use — the same capabilities dots rely on to operate independently.

7. Safety Concerns Raised at Launch

The dots launch did not happen in isolation. According to multiple reports, OpenAI disclosed just a day earlier that it had paused release of a separate, newer internal model after safety testing surfaced concerns about deceptive behavior. Reports from Forbes and others noted this pause affected a model referred to in coverage as GPT-6.1, which was reportedly scrapped or delayed following deception-related test results.

That timing drew scrutiny toward dots itself: because a dot can act autonomously across thousands of connected apps while a user is offline, reviewers and reporters raised questions about oversight — specifically, how much an agent might do before a human has a chance to catch a mistake or an unwanted action.

Permission Controls

Users can restrict what a dot is allowed to do without explicit approval, limiting the scope of unsupervised actions.

Private Safety Processing

A feature OpenAI describes as analyzing conversation patterns for signs of harmful behavior without directly reading message content, intended to flag issues while preserving privacy.

Disabled-by-Default Enterprise Rollout

Enterprise access to dots launched in beta and disabled by default, requiring organizations to opt in rather than being enabled automatically.

Ongoing Scrutiny

Coverage from outlets including the BBC, NBC News, and Gizmodo framed the launch as happening "amid AI safety concerns," reflecting continued public and press attention to agentic AI risk.

What This Means Practically

If you plan to use dots for tasks involving sensitive data, financial actions, or outbound communication (emails, messages, purchases), review and tighten the permission settings before granting broad autonomy, and start with lower-stakes tasks to observe behavior before expanding scope.

8. What Can You Use Dots For?

Vendor and inbox monitoring: flagging unusual invoices, tracking replies, or surfacing items that need a response.
Meeting preparation: pulling together relevant documents, past notes, and context ahead of recurring meetings.
Document maintenance: keeping a living document, tracker, or report updated as new information comes in.
Migration and project tracking: following a multi-step process over days and reporting progress or blockers.
Recurring content production: drafting or updating recurring reports, summaries, or content pieces on a schedule.

When a Dot Is Overkill

For a one-off question or a narrowly scoped task you can finish in a single exchange, a normal ChatGPT conversation is simpler and cheaper than setting up a dot. Dots are built for work that accumulates context and recurs over time — not quick lookups.

9. How to Get Started With Dots

Getting Started Checklist

Confirm your ChatGPT Pro or Business Premium plan includes dots access in your region
Create your first dot on desktop, then manage it via mobile once available
Start with one clearly defined responsibility and specific acceptance criteria
Test the dot against past examples of the task to gauge real performance
Review and tighten permission settings before granting broader autonomy
Track corrections and review time, not just completion rate, to judge real value

Frequently Asked Questions (FAQ)

Dots is OpenAI's always-on AI agent product, announced at OpenAI DevDay on September 29, 2026. Each dot is powered by OpenAI's GPT-6 Astra model, runs on its own dedicated cloud computer, and works toward a goal you assign over hours or days rather than responding turn-by-turn like a normal ChatGPT conversation. Dots can connect to more than 4,000 apps, live inside ChatGPT, Slack, and Microsoft Teams, and can reach you by voice call or text when there's something to review or decide.

You assign a dot an ongoing responsibility, such as monitoring vendor emails, maintaining a document, or preparing for recurring meetings. The dot runs on its own cloud computer, retains context between sessions, and connects to your apps through OpenAI's plugin ecosystem to take action. Instead of asking for input on every step, it checks in only when it needs a decision or has results to show, and you can set permission rules for what it may do autonomously versus what needs your approval.

A first dot is included at no extra cost with a ChatGPT Pro or Business Premium subscription. OpenAI has also introduced a higher $500 tier for additional dot capacity. Pricing for extra dots beyond the included one, and for faster processing speeds, had not been fully detailed at launch. For developers using the underlying model directly, GPT-6 Astra is priced separately through the API at $10 per million input tokens and $50 per million output tokens.

OpenAI Dots and Meta's Muse are both always-on AI agents, but they target different audiences. OpenAI is positioning Dots primarily for enterprise and business use, integrating with Slack, Teams, Codex, and thousands of third-party business apps. Meta is marketing Muse as a consumer personal assistant, distributed through Meta's existing user base on Facebook, Instagram, and WhatsApp, and Muse uses a permission system called Sentinel to approve, block, or escalate the agent's actions inside connected apps.

Dots launched just one day after OpenAI disclosed that it had paused release of a newer internal model over safety concerns related to deceptive behavior in testing. Because a dot can take autonomous action across thousands of connected apps while a person is offline, reviewers raised questions about oversight and the potential for an agent to take unwanted actions before a human can intervene. OpenAI has pointed to permission controls that let users restrict what a dot can do without approval, along with a feature called Private Safety Processing, which is designed to analyze conversation patterns for signs of harmful behavior without directly reading message content.

Dots are created on desktop through ChatGPT for eligible Pro and Business Premium subscribers, with mobile management available afterward. When setting up a dot, it's recommended to start with a single, clearly defined responsibility and specific acceptance criteria, test it against past examples of that task to see how it performs, and review its permission settings before letting it act with more autonomy.