What Is Dots by ChatGPT? OpenAI's Always-On AI Agents, Explained

The short answer

Dots are always-on AI agents inside ChatGPT. You give one a goal, connect the apps it needs, set what it can do on its own, and it keeps working between your conversations.

OpenAI announced them on 29 September 2026 at DevDay. Three things separate a dot from a normal ChatGPT chat:

1. It carries on without you. A chat waits for your next message. A dot keeps making progress toward the goal and comes back when it needs a decision.
2. It has its own computer. Each dot runs on GPT-6 Astra with its own cloud computer and browser, and reaches over 4,000 apps through plugins.
3. It works under rules. You decide which actions it takes alone, which need your approval, and which stay with you entirely.

A dot is pointed at your work: your inbox, your calendar, your projects, your code. It is not something you hand to a customer, because there is no public link to give them. That outward-facing job is a different category. A customer-facing AI agent built with Agentplace gets its own web page and URL, and your customers open it themselves to get a result.

Why did OpenAI build Dots?

The limitation dots address is simple. A chat session is a good thinker and a poor colleague. It answers well, then stops, and the next conversation starts from nothing.

That shape costs you in specific ways:

• You have to remember to ask. Nothing happens while you are in a meeting.
• You re-explain context every time, because the thread has no standing brief.
• Work spanning days gets split across threads you then reconcile by hand.
• A task needing five tool calls and one judgment call never gets delegated at all.

“This is the real-deal version of AI that we’ve always imagined for ourselves. It’s like an AI helper that always has your back, inspired by the cool versions of what we all watched in movies growing up.” Sam Altman, CEO, OpenAI, speaking at DevDay 2026

OpenAI is not alone. Meta released Muse in September, Google shipped Gemini Spark in May, and the startup Instinct is chasing the same idea. The category forming around them is the persistent agent: one that holds a goal over time rather than answering a single question well.

How does a dot work, step by step?

What it is	An always-on AI agent inside ChatGPT
Announced	29 September 2026, at OpenAI DevDay
Model	GPT-6 Astra
Where it runs	Its own cloud computer and browser
Apps	Over 4,000, through plugins
Channels	ChatGPT desktop, web and mobile, plus Slack and Teams. Texting is a limited beta
Plans	First dot included at no extra cost on Pro and Business Premium. Enterprise, Edu and Healthcare get a beta an admin must switch on
Markets	Pro excludes the EEA, Switzerland and the UK. Business Premium covers all supported ChatGPT regions
Setup	ChatGPT desktop app or desktop web. You cannot create a dot on mobile

Details from OpenAI’s announcement and its getting started guide, checked 30 September 2026.

1. Create it on desktop and give it a name

You make your first dot in the ChatGPT desktop app or on desktop web. Mobile can talk to a dot once it exists, but cannot create one.

Its default handle is @yourname-dot, and naming it changes the handle to match. You also pick a character, and it may generate itself a pet.

2. Connect the apps it needs

Plugins are how a dot gets hands. Connecting one gives it access to that app’s information for your tasks, authorised by you through your existing ChatGPT app controls.

Example: connect Gmail, Calendar and a project tool, and a dot can read the week, spot the conflict and draft the reschedule.

This is the same pattern behind MCP and business tools: the agent does not guess at your data, it acts through a connection you approved.

3. Set what it can do on its own

Custom Rules are the part worth reading carefully. For any action you describe, you choose one of four behaviours:

• Take action without asking.
• Take action if pre-approved, meaning you asked for it explicitly in your prompt.
• Ask before taking action.
• Hand off to you.

Underneath that, auto-review checks anything touching your accounts or sharing information against your instructions, your rules and OpenAI’s safety requirements.

Some things never move, whatever you set. Changing a password always stays with you.

4. Let it work from its own computer

Each dot gets a cloud computer and browser of its own, and you can open that computer at any time to watch what it is doing.

Your own machine stays separate by default. Connect it through the desktop app and the dot can work with local files, use local skills and fall back to your browser when its own is blocked. Revoking access is one confirmation.

For signing into supported sites, a dot can use saved passwords without those passwords being exposed to the model.

5. Let it keep going between conversations

Two things happen when you are not typing.

• Scheduled work. Reminders and recurring checks you asked for, managed under Scheduled in the dot’s profile.
• Proactive research. The dot looks for ways to help using connected apps, with tools restricted to read-only.

That read-only boundary is the important detail: during proactive research a dot cannot send messages, change app content, or control your browser or computer.

6. Review what comes back

Activity sits in the dot’s profile under In progress, Scheduled and Completed, and the dot messages you with progress and decisions that need you.

Pause stops it. Reset deletes its conversations, memories and scheduled tasks. Disconnecting an app does not delete what the dot already learned from it.

What can Dots do, and where do they stop?

What it does	Where it stops
Works toward a goal between conversations	OpenAI’s own caveat is that dots can still make mistakes, so review consequential work
Runs on its own cloud computer and browser	Your computer stays separate unless you connect it, and that starts off
Researches in the background	Read-only in proactive research. No messages sent, no app content changed
Connects to over 4,000 apps	Enabling dots does not grant access to every app or site, and disconnecting one does not erase what it learned
Reaches you in ChatGPT, Slack and Teams	It cannot call you at launch. Texting is a limited beta for Pro users in the US only
Uses your connected email account	It has no standalone email address of its own yet
Acts under Custom Rules you write	Sensitive tasks such as password changes always stay with you
Serves the person who owns it	Your customers cannot open your dot
Specialist dots with their own identity and credentials	In focused enterprise pilots, arranged with OpenAI’s engineering teams, not generally available
Customer-facing AI agent built with Agentplace	Works for your customers on its own page and URL. It does not know live availability or pricing unless you connect the system holding it

Two of those rows deserve more than a line.

Specialist dots are the organisational version, and they are early. OpenAI previewed dots that a company provisions with their own identity, credentials and access to systems of record, built on internal testing across procurement, invoice processing, email marketing, customer support and commercial contracting. It is starting as focused pilots, with a planned Microsoft Agent 365 integration for governance. That is a real direction, but it is a sales conversation today, not a signup.

Availability is narrower than the headlines suggest. Pro access skips the EEA, Switzerland and the UK, Enterprise and Edu need an admin to switch the beta on, and rollout is gradual even inside eligible markets.

How is Agentplace similar to Dots?

Agentplace creates AI agents that serve customers and help run a business. Each agent gets its own web page and URL. It uses business knowledge and authorized tools to deliver a personal plan, estimate, product comparison, document, or other result. The AI controls the interface and updates the work as customers make requests by text or voice.

Read that next to how a dot is built and the overlap runs deep. Both are agents with a persistent identity, their own environment, credentials you authorised, rules about when to ask a human, and work that runs when nobody is in the conversation.

What the agent has	In a dot	In an Agentplace agent
Its own environment	A cloud computer and browser	A web page and URL, with an interface that changes with the request
Connected apps	Over 4,000 through plugins, authorised by you	Composio’s catalogue of 1,000+ integrations and compatible MCP tools, authorised by the owner
Permission levels	Custom Rules: act alone, act if pre-approved, ask first, or hand off	Handoffs and approvals: you define which decisions require a person
Work with no prompt	Scheduled and recurring tasks, plus read-only proactive research	Scheduled tasks and events, each with a trigger, permitted actions and stop conditions
Memory	Inherits ChatGPT memories and forms its own, including from connected apps	Retains work and decisions, through a configured session and access flow
Human review	Auto-review checks actions before they proceed	Owners review changes before releasing them to customers
Who it serves	You, the account holder	Your customers

The last row is the whole difference. A dot and an Agentplace agent are the same generation of technology aimed at opposite sides of a business.

A dot needs your ChatGPT seat to reach it. That is right for your own work and wrong for a customer, who will not sign into your account to ask whether you cover their postcode.

An Agentplace agent runs the other direction. Its URL goes on your site, in an email or on a card. The visitor describes what they need in their own words, and the agent answers from your policies and catalogue, asks the questions that change the recommendation, and renders the result as a plan, estimate or comparison rather than a wall of text.

In practice, it works like this:

1. A visitor opens the agent’s page and describes a situation, by text or voice.
2. The agent asks only the questions that change the answer.
3. It applies your business knowledge: services, pricing rules, policies, service areas and qualification criteria.
4. It returns a personal result, and revises it when the customer changes something.
5. It acts through your connected tools, then hands off to a person where you decided a person is required.

You describe the outcome you want in plain language, and the AI Builder builds the agent. It can discuss the idea first, ask for the details it is missing, work from your existing website or uploaded materials, and test its own work before you see it. You choose when to publish.

What stays where it is:

• Live availability, inventory and current pricing come from the system that holds them, through a connection you set up.
• Your booking platform, CRM and accounting system keep doing their jobs. The agent connects to them.
• Decisions you marked as human stay human.

Agentplace has a free plan and paid plans that use AI credits. A credit is not a message, and usage depends on the model and the work performed, so check the current pricing page for amounts and limits.

To see the shape of it, the needs assessment agent and the B2B lead qualification agent are the closest starting points to this article’s use case.

When don’t you need a dot or a customer-facing agent?

• Is the work yours rather than your customers’? Then a dot, or a simpler assistant, is the right category. Do not build a customer-facing agent to read your own inbox.
• Do your customers already know exactly what they want? A booking page or a checkout is faster than a conversation, and cheaper to run.
• Is your question answered by a single fixed page? A well-written FAQ or a support chatbot may be enough. Our chatbot comparison covers where that line sits.
• Are you in the EEA, Switzerland or the UK on Pro? Dots are not available to you yet, so plan around that rather than waiting.
• Do you need one system of record? That is what your CRM or booking platform is for. An agent works with it, not instead of it.
• Is nothing about the task repeatable? One unusual job a quarter does not justify the setup.

Quick checklist

• Decide whose work you are automating: yours, or your customers’.
• For your own work, check you are on an eligible plan and in an eligible market.
• Write the rules before connecting apps, so you know what acts alone.
• List what the agent must never do without a person.
• Separate static knowledge, such as policies and catalogues, from live data such as availability.
• Name the systems that stay in place, and the connection each one needs.
• Decide how you will review consequential work, whoever the agent serves.

Related reading: What is an AI employee? · How to build an AI agent for your business · Agent skills · MCP and business tools

Bottom line

Dots and customer-facing agents are the same idea pointed in opposite directions: a dot does your work inside your ChatGPT account, while a customer-facing AI agent built with Agentplace does your customers’ work on a page you can link to.

• If you want an agent working your inbox, calendar, code and projects, and you are on ChatGPT Pro or Business Premium in an eligible market, create a dot.
• If you want an agent your customers open themselves, that asks the questions that change the recommendation, answers from your business knowledge, returns a personal result and books or hands off through the systems you already run, that is what Agentplace is designed for.