OpenAI dots carry on long after you close the window, checking on projects and chasing open items while your attention sits somewhere else.
This kind of always-on help redefines what a digital assistant means for your workday. You hand off ongoing responsibilities to a helper that follows through on its own and checks in when something needs your call.
With OpenAI dots, you get artificial intelligence (AI) agents that stay switched on, learn your priorities and act for you at any hour. Every dot operates from a private cloud workstation, browser included. Plug-ins link a dot to upward of 4,000 apps, so it can run multiple jobs side by side.
Three main channels connect you with your dot:
Context follows your dot across every channel. Ask it for a status update in Slack at 9 a.m., and you can pick up the same thread on your phone over lunch. A small texting beta also exists, open only to select Pro subscribers in the United States.
In ChatGPT Space, the shared workspace that replaced the Library, teammates can tag your dot on a page and ask it to keep the content current. The work then moves forward between meetings.
Bubbly cartoon characters give each dot a face. Launch designs come in pink, blue, green, yellow and purple. Some sport add-ons like glasses, headphones, berets and bow ties. The team behind them built a mix-and-match system of shapes and accessories to keep the look approachable without distracting from work.
Characters stay optional, too. You can pick a pet, let your dot generate one or skip the visuals entirely. Either way, small animations show your dot at work, which helps when several agents share one workspace.
GPT-6 Astra drives a dot’s independence, and OpenAI bills it as the world’s most intelligent and aligned model. The company also claims state-of-the-art results in computer use, browsing and software engineering. An agent leans on those skills to finish multistep work on a real desktop.
Picture a busy Monday. Your dot could update last week’s customer records, untangle two meetings that overlap and test a new software release before you’ve finished your coffee.
Apart from independence, Astra stands out most in speed. It reportedly completes Financial Modeling World Cup challenges about four times faster than the winning human competitor by operating Excel the way a person would. Range and speed together let a dot handle real assignments with little hand-holding.
As work spreads across more apps and tabs, the manual handoff between an AI tool’s answer and the finished task has become the As work spreads across more apps and tabs, the manual handoff between an AI tool’s answer and the finished task has become the bottleneck. Agents aim to close that gap by seeing jobs through on their own. Forecasts already put a timeline on that shift, projecting that one-third of generative AI interactions will use action models and autonomous agents by 2028.
Plenty of people have a head start. In September 2025, 21% of American workers said they do at least part of their job with AI. About a year earlier, the share sat at 16%.
The forecast shows where interactions are going, while the survey captures habits people have formed. A dot lands right where those trends meet, giving workers who regularly lean on AI a way to delegate whole tasks.
Getting a dot running happens inside OpenAI’s existing products. A little planning about what you want to hand off can make the first few days smoother.
Your dot starts in the ChatGPT desktop app, on the web or in the latest version of its mobile app. Setup requires an installed client or a browser on the computer. During onboarding, you’ll pick a name to replace the default @yourname-dot handle and choose how your dot looks.
Next, connect the apps you want your dot to use. The mobile version keeps these under Customize and then Plug-ins.
Messaging channels work differently because Slack and texting setups require a computer. Either way, you decide which services your dot reaches, and it operates only within the ones you approve.
Give your dot a first task with a clear goal, like drafting follow-ups to a client proposal. If you subscribe to Pro or Business Premium, one dot is included in your plan, and chatting with it doesn’t count against your regular ChatGPT allowance.
Jobs it kicks off inside Codex or ChatGPT Work drawdown that allowance the normal way. Launch month came with a more generous cap, and OpenAI has promised plan-by-plan terms once that window closes. If you expect to delegate heavily, watch those task limits, since conversation stays free while OpenAI meters the actual work.
Eligibility varies by plan and location, and Pro access covers users 18 and older outside the European Economic Area, Switzerland and the United Kingdom. Enterprise, Edu and Healthcare teams need an admin to switch on the beta first. Because rollout happens gradually, your dot may take several days to appear.
OpenAI pairs every connection with controls, and knowing how they work helps you decide what to share.
Goals, files, photos and context you add mid-task all shape what your dot learns. Apps you connect through plug-ins, like your personal email, feed it too. From day one, it inherits what ChatGPT remembers about you, plus your recent chats, then keeps adding notes from services it scans in the background.
For deeper reach, you can let your dot use your own computer, an option that stays off by default. Local access lets it open files, run commands and fall back on your personal browser when the cloud version hits a block. During supported sign-ins, your dot logs in with stored credentials, and the model never sees them.
Memory makes a dot more useful over time, and your dot holds onto chat and app history for its entire lifespan. Passwords, pictures and screen captures never make it into that record, while OpenAI encrypts your content in storage and in transit.
You can’t yet view, edit or delete specific saved details. In practice, anything a dot picks up while reviewing your inbox in the background stays with it until you reset the dot.
Unlinking an app or switching off ChatGPT Memory blocks anything new from flowing in, yet whatever your dot collected earlier remains.
Clearing everything takes a reset, which erases the dot’s chat history, notes and scheduled jobs in one go. Export any results you want to keep beforehand. Items your dot produced in Codex, ChatGPT or your file storage live elsewhere, so wiping the dot leaves them alone.
A single toggle, “Improve the model for everyone,” governs whether OpenAI learns from your dot’s chats, actions and linked-app data on personal plans. Company accounts on Business, Enterprise or Edu start out as opted out, and a dot’s proactive background research gets similar protection. Insights from that research can still reach training if they surface in a qualifying chat.
Three controls shape what your dot does without you:
An auto-review system checks riskier actions against your instructions before they run, like confirming the recipient and content of an outgoing email. Astra’s stronger alignment and safety testing show it resists jailbreaks and prompt injections significantly better than GPT-5.6 Sol, though it falls short of a guarantee.
Protection efforts also extend beyond any single company. The National Institute of Standards and Technology brings industry together on new AI agent security standards and open protocols.
Prompt injection works by planting secret commands in something your agent reads, like a site, a message or a shared file, hoping it will obey. Exposure is high, as security audits show that 73% of AI systems tested have vulnerabilities to these attacks.
Imagine a vendor email with hidden text that tells your dot to forward last quarter’s invoices to a stranger. With your inbox, calendar and work apps connected, an injection like that could leak data or trigger actions you never requested.
OpenAI’s monitoring can pause or stop a dot when it detects a safety concern, but important output still deserves a human check. Each new connection expands what your dot can accomplish and what an attacker could reach.
dots, Muse and a personal assistant each balance availability, judgment and cost differently.
| AI Assistants | OpenAI Dots | Meta Muse | Human Assistant |
| Availability | Always on via Slack, Microsoft Teams and ChatGPT | Runs in the background via the Muse app, web and WhatsApp | Set working hours |
| Judgment and relationships | Adapts to feedback | Picks up habits from chats | Reads people and builds rapport |
| Cost | First dot included with ChatGPT Pro or Business Premium | Free with limited tokens. Paid tiers cost $20 to $100 per month | Salary or hourly pay |
Under the hood, both companies park their agents on private virtual computers. Meta equips each Muse with its own persistent Linux computer and a full browser, then links it to email, calendar and Instagram. You can reach it on iOS, Android, Mac and WhatsApp.
Meta’s paid Power and Maximum plans raise the usage cap. Muse aims at every part of your life, while OpenAI builds dots around work tools like Slack and Microsoft Teams. That work-first design gives dots an edge at the office, inside software your colleagues already rely on.
A dot pulls ahead on availability, staying active through the night and keeping parallel threads moving before picking up where it left off. A human assistant counters with judgment built on lived experience, from sensing a frustrated client to knowing which relationships need attention. AI can pull information in seconds, but a person arguably knows best how to read the mood behind a terse email.
Specialized tools cover narrower jobs, too. Motion, ClickUp AI and TextCortex rank among the top AI personal assistants for work, handling scheduling, project management and writing. The right choice depends on whether your bottleneck involves hours in the day or judgment calls that still benefit from a person.
OpenAI dots turn the idea of a work assistant into something that owns ongoing responsibilities. That handoff requires real trust, and OpenAI’s controls let you build it one app and one permission at a time.
As your dot proves itself, Custom Rules let you widen what it handles on its own, while Activity View keeps its progress visible. Start small, then let the results decide how much of your to-do list it takes on.
