OpenAI launches Dots, permanent agents capable of continuing to work between two conversations, using their own cloud computer and connecting to user-authorized applications. Newness is not simply a more powerful ChatGPT. It transforms the AI of a tool that is requested on an ad hoc basis into a software collaborator to which a responsibility can be entrusted.
For a creator, consultant or small business, this can mean monitoring the arrival of new information, preparing content, keeping a project up to date or organising repetitive tasks while the user does something else. But this permanence also changes the nature of the risk: a bad statement or too wide permission can produce effects long after the first message.
OpenAI announced Dots the 29 september 2026. Deployment is progressive and varies according to package and region. To 30 september, Pro users located in the European Economic Area, Switzerland and the United Kingdom are not included. Business Premium users can access it in supported ChatGPT regions, while Enterprise, Edu and Healthcare spaces have a default beta disabled that the administrator must activate.
What the Dots of OpenAI really are
A Dot is an "always active" agent embedded in ChatGPT. It is powered by GPT-6 Astra, owns its own computer in the cloud and maintains the necessary context to advance a work over time.
It can receive a goal, use a browser, create files, run tools and work with the apps the user has chosen to connect. OpenAI announces an ecosystem of more than 4 000 applications accessible through its plugins.
The user can follow his activity in ChatGPT, talk to him on the Web, in the desktop application or on mobile after the initial configuration. OpenAI also announces exchanges from Slack and Teams. The Dot can send news about the promotion, ask for a decision or point out that a point requires validation.
This continuity is the essential difference with a classical conversation. In an ordinary chat, the user makes a request, waits for an answer and eventually returns later. A Dot can retain responsibility, monitor the evolution of a project, perform planned tasks and continue to progress between exchanges.
However, we must avoid exaggeration: "working 24 h/24" means neither perfect intelligence nor unlimited autonomy. The agent remains subject to package allowances, permissions granted, security rules and required validations. OpenAI also indicates that the features are still being phased out and may take several days to appear on eligible accounts.
Real change: moving from prompt to responsible
Since the arrival of the general assistants, much of the speech has focused on the art of writing the best prompt. This skill remains useful, but Dots moves the problem.
To delegate lasting responsibility, it is no longer enough to write a brilliant instruction. It is necessary to define:
A mission like "help me with my marketing" is too vague. A responsibility such as "every Monday, analyzes recurring questions received the previous week, proposes three topics and prepares drafts without publishing anything" has a perimeter, a pace, sources and a limit.
This difference is similar to the difference between a tool and a workstation. When the AI becomes permanent, the quality of the process counts more than the formula used in the first message.
Two practical uses for creators and independents
Creator transforms content without automatically publishing
Imagine a creator who publishes a long interview every week. His Dot can monitor the arrival of the transcription in an approved file, identify the most useful passages, prepare descriptions, propose several excerpts and write a first draft newsletter.
The agent gradually learns the corrections of the creator: length of hooks, subjects to avoid, preferred vocabulary or expected level of detail. The creator does not necessarily repeat all the explanation for each episode.
The good architecture, however, is to keep a validation before publication. An extract from its context, an imprecise quotation or an excessive commercial promise can undermine the professional's credibility.
Once the content is validated, a LinkHub VIFLY can bring together the episode, the resources cited and the associated offer. The Dot accelerates preparation; The central space makes the route understandable to the public.
Consultant follows his prospects and prepares appointments
Let's take a consultant who receives requests from several channels. A Dot could consult authorized applications, gather useful information, report unanswered requests and prepare a synthesis before each appointment.
It can also offer a follow-up email or remind that a proposal is coming to an end. But it should not invent a remission, accept a contractual clause or transmit confidential information to an unplanned recipient.
The course can end on VIFLY Booking, where the prospect chooses a clearly defined service and niche. Automation prepares and fluidizes the relationship; It must not make the offer more ambiguous.
In both examples, the value does not come from a maximum agent activity. It comes from the removal of precise repetitive work, without delegating decisions that engage the image, money or responsibility of the professional.
Permissions, Memory and Validations: The Limits to Understand
A permanent agent needs context to be useful. This is precisely what makes its setting sensitive.
OpenAI explains that each Dot works in an isolated cloud environment. Access to the personal computer is optional and disabled by default. The user chooses the connected applications and can view the current activity.
For proactive background searches, OpenAI indicates using tools limited to reading alone. These tasks cannot directly send a message, change the content of an application or control a browser. Subsequent action remains subject to ordinary rules.
The Custom Rules allow you to choose different behaviors: act without asking, act when a pre-authorised instruction, ask for confirmation or transmit the step to the user. A separate system called Auto-review examines certain actions before they are executed.
Sensitive transactions retain mandatory restrictions. Procurement requires approval. The definitive deletion of data or installation of software from an unrecognised source must be confirmed. Password change or money transfer remains the responsibility of the user.
These protections do not transform the Dot into an infallible system. OpenAI explicitly specifies that it can make mistakes. A page, document or email may also contain a malicious instruction to divert the agent, an attack called rapid injection.
Another important shade: disconnecting an application blocks new accesses, but does not automatically remove what the Dot has already built into its context. Documentation indicates that a reset removes the Dot, its conversations, memories and planned tasks.
For Business, Enterprise and Edu spaces, OpenAI states that it does not use content to drive its default models. On personal packages, use depends on the "Improving the Model for All" setting.
Availability: Dots are not yet accessible to all in France
Availability requires a clear distinction between packages.
The first Dot is included without charge in the relevant Pro and Business Premium packages. A detailed work allowance is included, with extended limits during the first month. OpenAI then plans to allow the addition of several Dots and the purchase of additional capacity, without yet detailing all future tariffs in the announcement.
The initial creation is done from ChatGPT on the desktop web or in the desktop application. Mobile use occurs after this configuration, but mobile creation and access from the mobile web are not supported at launch.
This situation can change rapidly. Readers must check the official documentation before signing up for a package only for this feature.
What to do now? The action plan VIFLY
The good first use is not the most spectacular. It is the one that saves time while remaining easy to verify and cancel.
The Dots are an important step: we are gradually moving on to an AI that responds to an AI that retains responsibility. For small businesses, this can unlock a real leverage effect. But it rewards above all activities already able to explain clearly their offers, their rules and their limits.
The blog VIFLY so will continue to judge these tools on a simple criterion: do they really transform the visibility and time gained into opportunities, without removing the professional's control of his decisions?
OpenAI Dots FAQ
What is a Dot in ChatGPT?
A Dot is a permanent agent of OpenAI, powered by GPT-6 Astra, which has its own cloud computer and can pursue responsibilities between multiple conversations.
Are Dots available in France?
Not for all users. At 30 September 2026, the Pro deployment initially excludes the European Economic Area, Switzerland and the United Kingdom. Business Premium is announced in the supported ChatGPT regions. Enterprise, Edu and Healthcare have a beta that can be activated by the administrator.
Can a Dot work without ChatGPT remaining open?
Yes. His cloud computer and planned tasks allow him to continue some work in the background. Its activity remains limited by permissions, safety rules and package quotas.
Can he send emails or make purchases alone?
The rules vary according to action and authorizations. Shipments and shares are controlled by the instructions and the Auto-Review. OpenAI indicates that purchases require user approval.
Can we connect your own computer?
Yes, but this option is disabled by default and must be allowed from desktop application. The Dot can then access local files and tools as part of permissions granted.
Are the data used to drive the models?
OpenAI states that the contents of Business, Enterprise and Edu spaces are not used for default training. On personal packages, this depends on the setting chosen by the user.