What Are ChatGPT Dots? How OpenAI’s Always-On AI Assistants Work

Most AI chats wait for the next prompt. OpenAI’s dots are designed to keep working after the conversation ends. Announced on 29 September 2026, dots are named, proactive assistants inside ChatGPT. A dot can learn how you work, use its own cloud computer, connect to approved apps, manage several projects, and return when it needs a decision. OpenAI describes this as a shift from asking AI for isolated answers to giving an AI assistant ongoing responsibility. OpenAI styles the product name as dots in lower case. People are also likely to search for it as “ChatGPT Dots,” so this guide uses both forms. ChatGPT Dots in one minute Question Answer What is a dot? A proactive AI assistant that can take responsibility for ongoing work, not only answer prompts. What powers it? OpenAI says dots are powered by GPT-6 Astra. Where does it work? In ChatGPT on desktop, web, and mobile, and through Slack and Microsoft Teams. Text messaging is planned. Can it use other apps? Yes. OpenAI says its plugin ecosystem can connect dots to more than 4,000 apps, subject to your permissions. Does it have a computer? Yes. Each dot has its own cloud computer and browser. You may also permit it to use your personal computer. Can it work in the background? Yes. It can continue projects and conduct restricted, read-only proactive research when you are away. Who can access it at launch? OpenAI is rolling it out to Pro and Business Premium users in eligible markets. Enterprise, Edu, and Healthcare workspaces can try the beta when an administrator enables it. Does it act without limits? No. App permissions, Custom Rules, approvals, Auto-review, monitoring, and mandatory handoffs restrict what it can do. Source: OpenAI’s Introducing dots announcement. Watch OpenAI’s dots launch film This is the same Live from the Lab launch film featured in OpenAI’s announcement. Select the image to watch the official video on Vimeo. Watch OpenAI’s official Live from the Lab launch film introducing dots. What exactly is a ChatGPT dot? A dot is an AI assistant that you set up around your goals, preferences, standards, and connected tools. You can name it, talk to it, send it work, check what it is doing, and correct it as a project develops. The important difference is the time horizon. A normal AI request often begins with a prompt and ends with a response. A dot can keep a project active, respond to new information, perform work in connected systems, and bring back decisions for your approval. OpenAI says a dot can work on several projects at once. It can also keep working when you are not actively chatting with it. That makes it closer to a persistent agent than a chatbot session. This does not make a dot an independent employee. It still relies on the access you grant, the instructions you provide, the websites and tools it can reach, and the safeguards that govern sensitive actions. OpenAI also states that dots can make mistakes and that consequential work should be reviewed. How dots work The product combines five layers: a capable model, persistent context, a cloud computer, connected apps, and a control system. 1. GPT-6 Astra interprets the goal OpenAI says dots are powered by GPT-6 Astra. The model is intended to understand the user’s goal, work through ambiguity, stay within the requested scope, and ask focused questions when an answer could change the result. The model is only one part of the system. A long-running assistant also needs somewhere to work, access to information, and rules for what it may do. 2. Each dot has its own cloud computer A dot can browse the web, create and review files, run tools, and use a browser on its own cloud computer. OpenAI says this environment is separate from your computer unless you decide to connect your device. You can inspect the dot’s computer while it works. If you permit access to your laptop, it may also work with local files or tools. That connection remains subject to device permissions, sandboxing, and action checks. 3. Connected apps provide working context Dots can use apps connected through ChatGPT’s existing controls. OpenAI says the plugin ecosystem gives dots potential access to more than 4,000 apps. That access is not automatic. You choose which accounts and apps to connect. The dot can only use the permissions available through those connections. Disconnecting an app stops new access through that connection, though information already added to the dot’s context may remain until you reset or manage that context. 4. The dot learns how you judge good work As you work together, the dot can learn your preferences, goals, working style, and quality standards. This accumulated context is what allows it to revise a project consistently instead of treating every prompt as an unrelated request. For example, a content lead could teach a dot the approved brand voice, audience, source rules, review checklist, and preferred output formats. When a new interview transcript arrives, the dot could find useful moments, prepare show notes, and draft channel-specific posts for review. 5. Controls decide when work can continue Permissions, built-in action rules, Custom Rules, user approvals, and a separate Auto-review system determine whether a planned action can proceed. Some actions always require confirmation or a manual handoff. This design matters because an AI agent can affect real accounts, files, people, and money. A wrong answer in a chat is one kind of error. Sending the wrong file or changing the wrong record is a different level of risk. How dots differ from a regular ChatGPT conversation Area Regular conversation Dot experience Main pattern You ask, ChatGPT responds You assign ongoing responsibility and review progress Duration Usually centred on the current exchange or project Designed to continue across time and changing inputs Work environment Chat and the tools used for the current task A dedicated cloud computer, browser, tools, and connected apps Initiative Usually waits for your next request Can run restricted proactive research and bring useful findings to you Context Conversation and available memory or project context A named assistant that learns your preferences and standards over time Channels Primarily ChatGPT ChatGPT, Slack, Teams, and planned text messaging Oversight Tool permissions and task approvals App permissions, Custom Rules, Activity View, approvals, Auto-review, and monitoring This is not a statement that standard ChatGPT lacks tools or memory. The difference is how OpenAI packages those capabilities around a persistent assistant with ongoing responsibility. What can a dot do? OpenAI’s launch material focuses on work that changes over time and requires several connected steps. Turn product feedback into tested fixes A developer’s dot could monitor customer feedback for repeated issues, define a small bug fix, build and test it, and prepare a pull request with evidence for review. The human developer still reviews the proposed change before it moves forward. Revise a launch when the product changes A launch plan often spreads across briefs, email, social posts, design files, sales material, and internal documents. When the product scope changes, a dot could trace the effect across those materials and prepare a coordinated revision. OpenAI’s launch example: a dot revises campaign materials as product scope changes. Source: OpenAI. Update research as evidence changes A scientist’s dot could rerun an analysis when new data appears, examine an unexpected result, update figures, and flag parts of the explanation that need human review. The useful part is not one calculation. It is keeping the analysis and narrative aligned as the evidence changes. Keep a complex sales proposal current For an enterprise deal, a dot could compare customer requirements with product documentation, identify missing tests, build a proof of concept, update a proposal, and raise unresolved concerns. Commitments and customer communication can remain with the sales lead and technical owner. Prepare content across formats A creator’s dot could process an interview transcript, select potential clips, create show notes, and draft social posts. Feedback on one asset can inform the others, reducing the need to repeat the same direction in every tool. OpenAI’s content-production example. Source: OpenAI. These examples share one feature: the work is not finished in a single prompt. The dot follows a goal through several systems and returns reviewable progress. What is proactive research? When you are not actively working with it, a dot can look for useful information in connected sources. OpenAI calls this proactive research. This background mode is deliberately restricted. It uses read-only tools for the apps you have connected. OpenAI says proactive research cannot send messages, change content in connected apps, or control a browser or computer. It can collect information and save private notes for the dot. Any action based on those findings must pass through the normal rules and checks. Imagine that your travel plans change, a customer replies to an important thread, or campaign performance moves outside an agreed range. A dot may notice the change and prepare context before you ask. It still cannot use proactive research as a hidden route to take unrestricted action. Where can you talk to a dot? OpenAI says you can message or call dots through ChatGPT on desktop, web, and mobile. You can also reach them through Slack and Microsoft Teams. Text messaging is listed as a future channel. The same dot carries context across these channels. You might start a project in ChatGPT, continue the discussion with colleagues in Slack, and later ask for an update from your phone. Cross-channel context can reduce repeated briefing. It also increases the importance of clear permissions because the assistant may encounter information from several workspaces. How OpenAI handles permissions, approvals, and safety Dots can take actions, so their safety model covers more than the wording of an answer. OpenAI describes several layers. A protected workspace Each dot runs in its own cloud environment. Sandboxing restricts the code and tools available inside that environment. OpenAI says user environments are isolated from one another, and the systems that enforce core safeguards sit outside the workspace a dot can modify. Secure sign-in For supported websites, a dot can use a secure sign-in flow that keeps passwords outside the model’s conversation context. The model pauses while credentials are entered, or a separate encrypted credential service can supply a saved password without placing it in the prompt. This protection only applies to supported secure flows. A password pasted into an ordinary message or readable document may still be visible to the model. App permissions You choose which apps a dot can access through ChatGPT’s connection settings. This is the first boundary. A dot cannot responsibly perform work in a system it cannot reach. Custom Rules Custom Rules let you define standing preferences such as “never send an email,” “always ask before publishing,” or “do not share customer data outside the approved workspace.” These rules operate inside OpenAI’s mandatory safety requirements and cannot remove required confirmations. Auto-review before actions Before a dot performs actions such as sending an email or changing a file, a separate system called Auto-review checks the proposed step against your instructions, Custom Rules, and built-in safety requirements. If the action is allowed, the dot can continue. If it is blocked, the dot may ask for approval or more information, try a permitted alternative, hand the step back to you, or stop. A user approval cannot override OpenAI’s core safety requirements. Mandatory confirmations and handoffs OpenAI says some actions require confirmation each time, including permanent deletion, running software from an unrecognised source, or granting sensitive access. Other steps, such as changing a password or transferring money between financial accounts, must be completed by the user. Purchases made with a saved merchant card require approval. Activity View and monitoring Activity View lets you see ongoing and delegated work, add context, change direction, or stop a task. OpenAI also says it monitors active work for potentially harmful behaviour. A safety concern can pause or stop the dot’s work and surface a warning. The fuller explanation is available in OpenAI’s article on safety, security, and privacy for dots. What happens to your data? OpenAI states that content from ChatGPT Business, Enterprise, and Edu workspaces is not used to improve its models by default. On personal ChatGPT plans, the “Improve the model for everyone” setting controls whether eligible dot conversations and work can be used for model improvement. OpenAI says it does not train directly on proactive-research threads or a dot’s private notes. If information from those notes later enters an eligible conversation or task, the normal account setting may apply. Data rules can change by plan and region. Teams handling customer, health, financial, legal, or confidential business information should review the current workspace terms and settings before connecting those systems. Who can use ChatGPT Dots? At launch, OpenAI is beginning the rollout in eligible markets for: ChatGPT Pro users ChatGPT Business Premium users ChatGPT Enterprise users, including Edu and Healthcare, when a workspace administrator enables the beta OpenAI says the first dot is included in Pro and Business Premium at no extra cost. Plans include an allowance for deeper work, with higher limits during the first month after launch. The company has not presented one universal allowance in the announcement, so users should check the limits shown in their own account. OpenAI plans to let users add more dots later and increase either a dot’s speed or its monthly workload. How to create your first dot OpenAI’s current setup path is short: Open ChatGPT on a desktop browser or in the desktop app. Choose the option to create your first dot. Give the dot a name and define its main responsibility. Connect only the apps needed for that responsibility. Set boundaries for messages, publishing, file changes, purchases, and sensitive information. Give it a real task with a clear finish line. Inspect its progress and correct the output. Add Custom Rules when a preference should apply repeatedly. After setup, you can message the dot from the ChatGPT mobile app. A practical first brief For a marketing lead, a first task could look like this: Review the latest campaign brief, approved brand examples, and this week’s performance report. Prepare revised copy for the three underperforming ads. Explain what changed and why. Do not publish, change budgets, or message the client. Stop when the drafts and supporting evidence are ready for my review. This brief defines the outcome, sources, limits, and review point. That is safer and more useful than “manage my ads.” Where dots may fall short Dots extend what an AI assistant can do, but they do not remove the usual weaknesses of AI or automation. They can misunderstand the goal. A persistent agent can carry a wrong assumption further than a one-off chat response. Connected information can contain malicious instructions. OpenAI uses safeguards against prompt injection, but it does not claim the risk has disappeared. A tool may be unavailable or behave differently. Websites change, sessions expire, permissions vary, and integrations fail. Memory can become stale. Important facts should be checked against the current source before a consequential action. Approval fatigue is possible. Loose rules may create too many interruptions. Rules that are too broad may create unnecessary risk. Availability is limited. Dots are rolling out by plan and market, and some enterprise access depends on an administrator. Usage is not unlimited. Deeper work uses an allowance, even though ordinary conversations with a dot do not count toward ChatGPT message limits. The safest starting point is read, analyse, and draft. Add external actions only after the workflow produces reliable results and the approval boundary is clear. What dots change about everyday AI use The main change is not a better answer box. It is a different unit of work. Instead of asking for a campaign summary, you can assign responsibility for keeping the campaign brief, performance evidence, draft revisions, and review questions aligned. Instead of asking for one bug fix, you can ask an assistant to watch feedback, reproduce a problem, prepare a tested change, and bring the pull request back for review. That creates more leverage, but it also makes process design more important. A useful dot needs a defined job, trusted sources, narrow permissions, a clear output, and an explicit point where a human decides what happens next. Frequently asked questions Are ChatGPT Dots the same as custom GPTs? No. A custom GPT is a configured ChatGPT experience with instructions, knowledge, and selected capabilities. A dot is presented as an always-on assistant that can keep responsibility for ongoing work, use its own cloud computer, operate across channels, and proactively research within strict limits. Can a dot work when my computer is off? Yes. Its work runs on a cloud computer. If a task specifically needs your local device, the device must be connected and the required permission must be available. Can a dot send emails or publish content on its own? It may be able to perform those actions when the connected tool, your instructions, Custom Rules, approval state, and safety requirements allow it. You can require approval or block an action entirely. Some sensitive actions always need confirmation or a manual handoff. Can a dot use my passwords? For supported sites, secure sign-in and saved-password flows can authenticate the browser without exposing the password to the model. Do not paste passwords or one-time codes into an ordinary chat. Does proactive research let a dot change my apps in the background? No. OpenAI says proactive research uses read-only tools. It cannot directly send messages, change connected content, or control a browser or computer. Is ChatGPT Dots available in India? OpenAI says dots are rolling out in eligible markets, but the launch announcement does not provide a complete country list. Check your ChatGPT account for the create-a-dot option and the current eligibility page before assuming access. Will dots replace human review? They are designed to reduce repeated execution and preparation. OpenAI still tells users to review consequential work. Decisions involving customers, money, permissions, legal commitments, confidential data, or production systems should keep a clear human approval point. Sources OpenAI, Introducing dots, 29 September 2026. OpenAI, How we build safety, security, and privacy into dots, 29 September 2026. OpenAI, Getting started with your dot, accessed 1 October 2026.