--- title: "What Are ChatGPT Dots? OpenAI's Always-On Agent Explained" url: "https://goremy.ai/blog/articles/openai-chatgpt-dots-always-on-agent" author: Luis Chavez-Mattos editor: Luis Chavez-Mattos editor_title: Director, Product published: "2026-11-07T00:40:08.894Z" updated: "2026-11-10T17:11:31.510Z" topic: AI Agents & Automation description: ChatGPT Dots are agents with their own cloud computer that run in the background, use memory, connected apps and Codex, and take texts from your phone. --- # What Are ChatGPT Dots? OpenAI's Always-On Agent Explained > ChatGPT Dots are agents with their own cloud computer that run in the background, use memory, connected apps and Codex, and take texts from your phone. Edited by Luis Chavez-Mattos, Director, Product · Published 2026-11-07 · Updated 2026-11-10 ## What are ChatGPT Dots? A Dot is an OpenAI agent that lives inside ChatGPT and keeps working after you close the window. It runs on OpenAI's frontier models and has its own cloud computer, so it does not depend on your laptop being open. You hand it a job, it works on it in the background, and it contacts you only when it needs a decision or some input. You can also text it from your phone to keep things moving. ## TL;DR - A Dot is a **persistent agent with its own cloud computer**, built into ChatGPT and powered by OpenAI's family of frontier models. - You assign it a task and let it run, rather than **prompting it step by step** the way you would in a normal chat. - It draws on **ChatGPT memory and the apps you connect**, and it can use Codex for building things. - In one demonstrated example, a Dot **planned a product launch and built the landing page in Codex**, pinging the user only when it needed something. - You can **text your Dot from your phone** to nudge work along while away from your computer. - Pricing, rollout details, and usage limits were not part of the material available for this piece, so treat those as **open questions** until OpenAI documents them. ## How is a Dot different from a normal chat session? A normal ChatGPT session is a turn-based exchange. You write a prompt, the model answers, and the work effectively pauses until you write again. If you close the tab, nothing keeps happening. You are the scheduler, the project manager, and the person who decides what comes next. A Dot flips that arrangement. Three differences stand out. | | Normal chat session | Dot | |---|---|---| | Lifespan | Lasts as long as you are in the conversation | Described as an agent that "never logs off" | | Compute | Model responds to prompts | Has its own cloud computer | | Your role | Prompt after prompt | Hand off a task, answer only when pinged | | Context | What is in the thread, plus memory | ChatGPT memory plus connected apps | | Contact | You open ChatGPT | You can text it from your phone | The practical shift is from supervising to delegating. You stop babysitting the model with prompts and let it run a task to completion, checking in only when it asks. ## Why does a Dot need its own cloud computer? Long-running work needs somewhere to happen. A chat model can write code in a message, but an agent that builds a landing page, keeps files, and returns to the task hours later needs a persistent environment. A dedicated cloud machine gives the agent that environment, so you can hand it a task and let it run on its own. That is the core architectural idea behind a Dot. The computer is the agent's workspace, not yours. Your phone or browser becomes a way to steer it, not the place the work runs. ## How do Dots use memory, connected apps, and Codex? Per the description of the feature, a Dot pulls from ChatGPT memory and the apps you connect, with Codex also working in the background. - **ChatGPT memory.** The agent can use what ChatGPT already knows about you, your preferences, and your ongoing projects. That cuts down on re-explaining context for every task. - **Connected apps.** Whatever apps you hook up to ChatGPT are available to the Dot. This is what lets an agent act on real information instead of working from a blank slate. - **Codex.** OpenAI's coding agent does the building. Codex and the connected-app work both run in the background while you do something else. Together these give the agent knowledge of you, access to your tools, and the ability to produce working artifacts such as code and pages. Each piece existed in ChatGPT before. The Dot is the layer that ties them to a long-lived agent. ## What does a Dot look like in practice? The clearest example comes from a creator who tried the feature. They gave their Dot a product launch. The Dot mapped out the plan, built the landing page in Codex, and contacted them only when it needed something from them. That last detail matters most. The value of an always-on agent is not raw speed. It is the reduction in interruptions. A good delegate works through everything it can handle and escalates only the decisions that need you, such as approving copy, choosing between options, or supplying a missing credential or asset. Note the limits of this evidence. It is one person's account of one task. It does not tell you how reliably a Dot handles ambiguous work, how often it pings, or how it recovers from mistakes. ## How do you talk to a Dot when you are away from your computer? You text it. The feature lets you text your Dot straight from your phone to keep things moving while you are away from your computer. For anyone who has left a long agent run waiting on a single approval, this is a meaningful convenience. The agent's blocker becomes a text reply instead of a delayed return to your desk. ## What do we still not know about Dots? The source material is a short walkthrough, not documentation. Several things a builder would want to know are not covered: - Which plans or regions get access, and what Dots cost. - Which specific models run underneath, beyond "OpenAI's family of frontier models." - How much autonomy a Dot has, and what permissions it asks for before acting in connected apps. - How many Dots you can run at once, and any limits on runtime. - How a Dot handles failures, and what logs or review tools you get. Check OpenAI's official announcement and help pages before building a workflow around any of these. ## Should people who build with AI pay attention to Dots? Yes, mainly for what it signals. Agents are moving from something you invoke to something that persists. If you build products or internal automations, three questions follow from this design. 1. **Where does the agent run?** A dedicated cloud computer per agent is a different model from a tool call inside a chat turn. 2. **How does it ask for help?** Escalation design (when to interrupt a human, and through which channel) is becoming a core product decision. 3. **What context does it carry?** Memory plus connected apps makes an agent more useful and raises the stakes on permissions and data handling. Whether Dots deliver on this in daily use is still to be shown. The concept is clear, though, and it is a useful template for thinking about any background agent. ## Frequently Asked Questions ### Is a Dot just a ChatGPT chat that stays open? No. The key difference is that a Dot has its own cloud computer and keeps working without you present. A chat session only advances when you send a prompt. A Dot takes a task and runs it, contacting you when it needs input. ### Can a Dot build things, or does it only plan? According to the demonstration, it can do both. The Dot in the product launch example mapped the plan and built the landing page using Codex. Codex handles the coding side of the work in the background. ### How do I give a Dot instructions from my phone? You can text your Dot. This lets you keep a task moving when you are away from your computer, for example by answering a question the agent raised. OpenAI's own documentation should cover the exact setup. ### Does a Dot know who I am? It draws on your ChatGPT memory and the apps you connect, so it can start with context about you and your work. The extent of what it can access depends on what you have stored in memory and which apps you have linked. ### How much do Dots cost? The available material does not say. Pricing, plan eligibility, and usage limits would need to come from OpenAI directly.