OpenAI Dots: Always-On ChatGPT

Hi, this is Alex from DEV STORIES DOT EU. OpenAI Dots: Always-On ChatGPT, episode 1 of 6. What if your AI kept working while your laptop was closed? You ask for a complex data analysis, shut your machine, and head home. By the time you sit down the next morning, the task is finished and waiting for your approval to proceed. This is the exact capability provided by an OpenAI Dot. A dot is an always-on agent powered by GPT-6 Astra. Listeners might confuse this with a standard chat interface where you type a prompt and wait for a reply. A dot is fundamentally different. It lives in the cloud and has its own dedicated computer and web browser. Because it operates independently of your local hardware, you can reach it at any time, and it continues to work even when your computer is completely powered off. The primary role of a dot is to manage ongoing responsibilities. It works between your conversations, tracks progress, and figures out what needs to happen next. It decides on its own when to pause and when to wake up to continue working. You do not have to set up a fixed schedule for every follow-up. You simply give it a goal, and it follows through as circumstances change. It handles research, analyzes data, prepares documents, and builds software. Furthermore, it can use background agents to work on several things in parallel. While it executes these tasks, you can keep talking to it. It reaches out to you with results or when it hits a decision point that requires your input. Consider a scenario where you are managing an offsite event. You hand the coordination over to your dot. As people confirm their attendance, your dot updates the headcount on its own. It then prepares a request for revised pricing based on the new numbers. Rather than just firing off the message, it asks you for approval before sending that request. It manages the logistics in the background and surfaces the key decision to you. To do this effectively, your dot draws on a deep pool of context. It uses the active conversation, relevant ChatGPT memory, its own saved notes, and your personal preferences. This means you can return to the dot later with new information or entirely new priorities. The dot incorporates the changes and keeps moving forward. You never have to restart the work just because a parameter changed. Pay attention to how this fits into the broader OpenAI toolkit. Standard Chat is for answers, explanations, ideas, and drafting text. ChatGPT Work is designed for delegating a specific task, such as a focused piece of research or creating a file, either immediately or on a schedule. Codex is built strictly for writing, debugging, testing, and reviewing software. Your dot sits above these tools to keep track of a long-term responsibility. It handles the high-level orchestration, tracking next steps, and following up. When needed, your dot can actually delegate individual tasks directly to ChatGPT Work or Codex to get them done. Dots are rolling out gradually to eligible accounts. As with any AI system, your dot can make mistakes. You should always review important results and verify their sources before taking action. The defining shift here is moving from a tool you actively drive to an agent that manages its own momentum and asks for directions only when it reaches a crossroads. If you find these technical breakdowns helpful, you can support the show by searching for DevStoriesEU on Patreon. Thanks for listening, happy coding everyone!

Hi, this is Alex from DEV STORIES DOT EU. OpenAI Dots: Always-On ChatGPT, episode 2 of 6. Setting up a dot asks for very little: a name, a look, and the apps you want it to use. What makes it useful comes later, from the first piece of real work you hand it. Today, we set up your first dot. First, a quick check on access. Dots are rolling out gradually. On the Pro 100, Pro 200, and Pro 500 plans, they are for users over eighteen outside the European Economic Area, the United Kingdom, and Switzerland. Business Premium and Enterprise are rolling out worldwide, and on Enterprise, dots are off by default until a workspace administrator turns them on. You can create your dot in the ChatGPT mobile app, the desktop app, or a desktop browser. Mobile web is not supported, so on a phone, use the latest version of the ChatGPT app. Whichever device you start on, you continue with the same dot across your devices afterwards. Setup has three steps. First, open dots and follow the introduction. Second, give your dot a name and pick its appearance: its shape, color, eyes, glasses, and accessories. You can keep the defaults and change any of this later. The name matters a little more than the look, because it becomes part of your dot's handle. Every dot starts with a handle made of your name followed by the word dot. Naming it updates the handle, so if your name is Tibo and you name your dot Alfred, the handle becomes at tibo dash alfred. Third, connect plugins for the apps you want your dot to use, such as email, calendar, and files. This step is optional. You can skip it and add apps later. There is one more connection you can make, and it is optional too: your own computer. Your dot already has its own computer and browser in the cloud, so it can work without your device. Connecting your computer lets it use files on your machine, the context of your signed-in browser, or your local Codex tasks. You set that up from the desktop app, during setup or later, even if you created your dot on your phone. While your dot uses your computer, keep it online with the ChatGPT app open. These connections stay separate. Adding a messaging channel such as Slack does not give your dot your inbox, your other apps, or your computer. Each connection is its own decision. Then comes the first conversation. Your dot introduces itself, and uses the context and connected apps available to it to learn about your work and suggest where it can help. You can start talking while it gathers that context, and if it needs more information or more access, it can ask. It starts with relevant ChatGPT memory, and it keeps its own notes as you work together. The best way to begin is with work you want it to keep track of, rather than a one-off question. Say you own a customer proposal whose requirements keep changing. Describe what matters, share the draft and the customer's requirements, and explain which decisions need your input, such as anything that changes the price. Ask it to track the decisions, deadlines, and open questions, and to tell you when something needs your attention. Then review its first result and correct the details it missed. When it helps, your dot can hand parts of the work to ChatGPT Work or Codex. The setup takes a name and a few choices. The value comes from giving your dot one real responsibility, with the sources and the decision points spelled out. Thanks for hanging out. Hope you picked up something new.

Hi, this is Alex from DEV STORIES DOT EU. OpenAI Dots: Always-On ChatGPT, episode 3 of 6. You switch from a deep planning session on your laptop to a quick question in your team chat, and suddenly your AI assistant has forgotten everything you just discussed. You have to paste in old links and summarize your own project just to get back to where you started. That friction disappears when you use Talk to It Anywhere. You reach the same dot through each connected contact method. Switching between ChatGPT, Slack, and Microsoft Teams does not create a separate dot. It also does not reset its memory or its saved notes. You connect a contact method directly from your dot's profile. Connecting one does not give your dot access to your inbox, other apps, or your computer. You can also message your dot in Microsoft Teams, which the admin guide says is currently an invite-only alpha. Texting is coming soon and is not available yet. Messages stay in their own channels. Slack and Teams show only the messages exchanged there, and a message in ChatGPT is not copied into Slack. The data is isolated, but your dot can use relevant context across them to inform its answers. Think about a continuous product launch review. You forward a launch plan to your dot in a Slack direct message. The next morning, it flags a schedule risk based on what it read. You then review its draft follow-up right there in a thread. In Slack, you can send a direct message or add your dot to a channel and mention it in a thread. Adding it to a channel does not start monitoring. You have to explicitly say which channels to check, what to watch for, and when to notify you. You should ask it to confirm the recurring task so you know the instructions took hold. By default, your dot responds only to you in Slack. You can instruct it to engage with others, and existing permissions and safeguards still apply. Before sharing information from a private conversation with other people, it checks that you have allowed it. Your dot can message you privately to ask what you are comfortable sharing before it replies in a channel. You can manage the noise by telling it where to send different updates. For example, instruct it to send routine progress in ChatGPT and major decisions in Slack. Later, you call it from ChatGPT to talk through the open decisions while typing a detail into the chat. You call your dot with the phone button in its conversation in the ChatGPT mobile app, desktop app, or a desktop browser. Mobile web is not supported. Calls initiated by your dot are planned for after launch, so today you start the call. Listen closely to this dynamic. You can type messages while talking, and your dot can message you with progress or a question. Calls use selected conversation context, which can differ from what a background task has. Ending a call ends the voice conversation, but assigned work can continue running smoothly out of sight. Maintaining context across text chat, team workspaces, and voice calls allows a single delegated task to evolve naturally throughout your workday without constant repetition. That is all for this one. Thanks for listening, and keep building!

Hi, this is Alex from DEV STORIES DOT EU. OpenAI Dots: Always-On ChatGPT, episode 4 of 6. You ask your AI to pull a weekly report from a dashboard, but it fails because it cannot see your active login session. You are logged in, but your assistant is not. That happens because your AI operates in entirely separate environments, and managing those spaces is how you bridge the gap between Computers and Apps. When your dot works on a task, it defaults to its own built-in cloud computer. This machine is completely separate from your devices. It has its own file system, its own software, and its own web browser sessions. Because it lives in the cloud, it keeps its state between periods of use, and cloud work continues while your devices are off. However, because its browser is separate, signing in to a website on your computer does not sign your dot in. Let us say your dot is trying to pull that dashboard report, and the website hits it with a login screen. It cannot proceed alone. Instead, your dot sends you a sign-in request with a private form. You enter your credentials and any verification code there, and the form sends them to the remote browser outside your conversation. Once signed in, the session can stay available for later work until you sign out or the website expires it. Saving the login to Passwords is optional, and reusing a saved login for a new sign-in needs your confirmation. Sometimes a private form is not enough. You can open your dot's cloud computer to watch it, but opening it alone does not give you control. You take over its mouse and keyboard when you want to do a step yourself, for example to sign in on the website, and then hand control back so your dot can continue. Certain websites simply block cloud browsers or require extra verification. You can then ask your dot to try your connected computer, which may create a separate local task and does not transfer the cloud browser's session. You connect your own computer once, from the ChatGPT app on that computer, through your dot's profile. You can connect only one personal computer at a time. It must be online with the ChatGPT app open while your dot uses it. An Offline status means it is unavailable but does not revoke access, and you revoke access explicitly when you want to remove it. With it connected, your dot can use local files, code, and apps. Your dot can use supported plugins installed and enabled for your account, with their connected accounts and your existing permissions. For example, it can use Gmail to find relevant email or Google Drive to read and update documents. You can let it read email without letting it send email. Keep in mind that local skills require a connected computer. Adding a contact method does not grant access to apps or your computer, so messaging your dot in Slack does not connect your inbox. The power of your dot scales with its access, but understanding the hard boundary between its cloud workspace and your personal hardware is what keeps those workflows predictable. Thanks for spending a few minutes with me. Until next time, take it easy.

Hi, this is Alex from DEV STORIES DOT EU. OpenAI Dots: Always-On ChatGPT, episode 5 of 6. When a project's priorities change halfway through, you don't start over with your dot. You tell it what changed, in the same conversation, and the work already underway changes with it. Today is about how a dot keeps track of tasks over time, and what it remembers while it does. Let's follow one responsibility. Say you hand your dot the project checklist for a release that ships in four weeks, and you ask it to keep that checklist current. The first thing to know is that your dot keeps working between conversations. It decides when to pause and when to wake up to continue, so you don't have to put every follow-up on a fixed schedule. When the work is bigger than one thread, it can split it among background agents that run in parallel and report back to it, while you keep talking to it. Each new task it creates gets its own conversation, with the instructions and context your dot gives it for that work. It does not automatically receive every conversation you have had with your dot. Now say the release date moves up a week. You don't open a new chat. You tell your dot in the same conversation what to change in the work already underway, and it can check results and send follow-up instructions to the tasks it created. In the desktop app, your dot's Activity view shows each task's progress, files, results, and any requests for your input. One habit worth keeping: a completed run doesn't by itself confirm that the requested result was achieved, so review the output and any reported errors. Some of this work should happen on a clock. Checking the planning channel every weekday morning is a recurring task, and recurring tasks need a saved schedule. A good request says what to check or update, when to run, including the time zone and an end date, which changes deserve a notification, and where to send the results. For example: check the connected planning channel each weekday at 9 AM Pacific for the next four weeks, update the project checklist when a deadline changes, and message me in ChatGPT only if a deadline is at risk or you need a decision. Then ask it to confirm what it saved. Keep in mind that stopping active work and canceling a schedule are two separate actions. Other work should happen when something occurs. If a connected service supports event monitoring, you can ask your dot to respond to a specific event instead, for example, to investigate new bug reports in a connected Slack channel and prepare fixes for your review. Tell it which events to watch for and when to notify you, and ask it to confirm which events it can follow. Connecting Slack on its own does not create a monitoring task. Through all of this, your dot remembers in layers. It uses the context of the conversation it's working in. It starts with relevant ChatGPT memory. And it keeps its own notes about your preferences, decisions, and ongoing work. Those notes are separate from ChatGPT's saved memory, and they are not a full transcript. They also carry across ChatGPT, Slack, and Teams, but being able to use something is not permission to share it. A deadline you mentioned privately can help it prepare your update, yet sharing private details in a team channel still needs your permission. Finally, your dot can do proactive research on information it has permission to read, and keep private notes about what it finds. It might notice that a release decision conflicts with a draft you shared last week, and bring that to you as a suggestion. That research doesn't send messages or change your apps on its own. Any follow-up action still goes through the permissions and approvals for that work. One conversation, one dot, and work that keeps moving while you're elsewhere. That is all for this one. Thanks for listening, and keep building!

Hi, this is Alex from DEV STORIES DOT EU. OpenAI Dots: Always-On ChatGPT, episode 6 of 6. Giving an agent autonomy does not mean blindly handing over the keys. You might want your dot to draft customer replies all day, but you probably do not want it firing off emails without a final look. The answer is Staying in Control. Built-in safeguards and your existing app permissions apply from the start. Asking your dot to draft replies does not automatically give it permission to send them. Before your dot takes an action that could affect your accounts or share information, an automatic review takes place. This review checks the pending action against your instructions, your permissions, your custom rules, and built-in safety requirements. The system then decides whether the action can proceed, needs your approval, or is a step you must do yourself, such as changing a password. You do not need a custom rule for every single approval. You start by giving clear instructions in your conversation. Custom rules are optional controls you add for specific ongoing boundaries. Staying with the customer support scenario, you have four distinct ways to control how your dot behaves when it acts on those drafted replies. Two choices let your dot act. You can tell it to take action without asking, allowing it to work freely. Alternatively, you can tell it to take action when you say so. This means it proceeds when you explicitly request the action, and otherwise asks you immediately before acting. The other two choices keep you firmly in the loop. You can instruct the dot to ask before taking action, which perfectly suits reviewing and sending those drafted messages to customers. Or you can tell it to hand off to you. This prompts the dot to ask you to take the action yourself, which fits high-stakes operations like deleting shared project files. Custom rules are just instructions your dot tries to follow, and it can make mistakes. They do not grant access to apps or computers, and they do not override built-in safety requirements or required confirmations, such as needing approval to use a saved login. You review your dot's work and answer its requests for decisions or approvals in its Activity view. Scheduled tasks are located in Scheduled. If you need to halt operations, you use Pause. Pause stops your dot's current main task, but it does not stop every delegated task or cancel future scheduled runs. You must stop or disable those separately. Resume continues a paused dot. Stopping work never undoes actions already completed. Deleting your dot cannot be undone. Always save any results you need first and read the confirmation screen to understand how it handles conversations, memories, and scheduled tasks. Just like pausing, deleting your dot does not undo changes already made in connected apps, nor does it recall messages already delivered to other people. Your standard ChatGPT data controls apply to eligible conversations with your dot and the work it carries out. OpenAI does not train models directly on its proactive research or private notes. On a company workspace, admins decide who can use dots and what they may do. They control permissions for joining Slack, using a local computer, using custom rules, and using the cloud browser and plugin actions. On Enterprise plans, dots are off until an admin turns them on. During the Enterprise beta, dots do not support data residency or inference residency. This brings us to the end of our series. I encourage you to read the official documentation, try these controls hands-on, and visit devstories dot eu to suggest topics for future series. Thanks for spending a few minutes with me. Until next time, take it easy.