Back to the blog
Always-on AI agents: OpenAI's dots and how to trial them with care
AI & automation

Always-on AI agents: OpenAI's dots and how to trial them with care

On 29 September OpenAI introduced dots, always-on agents with their own cloud computer and access to more than 4,000 apps. In Italy they are available with Business Premium, while Pro is excluded at launch; try them first inside a limited perimeter.

8 October 2026· 7 min·by NaCode Studios
Share
In short

On 29 September OpenAI introduced dots, always-on agents with their own cloud computer and access to more than 4,000 apps. In Italy they are available with Business Premium, while Pro is excluded at launch; try them first inside a limited perimeter.

On 29 September 2026, at DevDay, OpenAI introduced dots: always-on agents inside ChatGPT, each with its own cloud computer and browser, able to connect to more than 4,000 apps and built on GPT-6 Astra. They keep working when you are not there, learn from your feedback and can be reached from ChatGPT, Slack or Teams. In Italy they are available with Business Premium; Pro subscribers in the European Economic Area are excluded at launch. Before turning one on, decide which accounts it will reach and who approves its actions.

Updated 8 October 2026. This text rests on OpenAI's pages, Help Center included, and on two news outlets, Fortune and CNBC. The Help Center was edited as recently as the last few hours, so check plans and countries in your own workspace before deciding. The notes on personal data are general information, not legal advice.

Availability and pricing

PlanDots at launch
ProYes, except the European Economic Area, Switzerland and the United Kingdom
Business PremiumYes, in all regions where ChatGPT is supported
Enterprise, Edu, HealthcareBeta, on only if the workspace admin enables it

Italy is on OpenAI's list of countries where ChatGPT is supported; access arrives gradually. A Premium seat in ChatGPT Business costs $125 per user per month, or $100 a month with annual billing. A workspace needs at least two seats, which can be one Standard ($25 a month) and one Premium: if the Premium seat is the one that dots require, the smallest trial costs $150 a month on monthly billing [illustrative estimate, to be confirmed against your own project]. On the annual plan the seats stay committed until the end of the year and, OpenAI writes, unused seats are generally not refundable.

The first dot is included in the plan at no extra cost. For the first month after launch, using dots does not count toward the plan allowances, and OpenAI will then share usage terms for each plan. More dots are planned for the future, and Fortune writes that their price has not been named.

Permissions and risks

According to OpenAI, changing a password or moving money between accounts stays with the user. Permanently deleting data or installing software can require your confirmation each time, and purchases with a card saved on a website need your approval, which you can give in advance. With Custom Rules you can allow specific actions, require confirmation for them or block them, and the core safety checks stay on in every case. Before actions such as sending an email, a separate system called Auto-review checks the recipient and the text.

When you are not using it, a dot does what OpenAI calls proactive research: it reads the connected apps with read-only tools that cannot send messages, change content or control the browser or the computer.

The risk of prompt injection remains: a website, an email or a document can carry instructions that push the dot to do something you did not ask for, such as sharing private data. According to OpenAI that content does not grant permission on its own, the protections reduce the risk without removing it, and dots can still make mistakes. A dot's memory cannot be viewed or corrected entry by entry, and disconnecting an app does not erase what the dot already holds in its context. Deleting the dot erases its context; the files, Codex threads and conversations it created stay stored separately.

Setting up a trial

  1. Pick a low-risk task with a measurable result, for example a weekly summary of the requests received or a check of deadlines.
  2. Look in ChatGPT's plugin settings at what is already connected, because a dot can use existing connections; then connect one app at a time and read the permissions it asks for.
  3. Use a mailbox and a calendar created for the trial, holding only the data it needs, and do not connect your own computer.
  4. Set the Custom Rules so the dot asks for confirmation before sending messages, sharing files or deleting.
  5. Give the seat to one person only and choose monthly billing. For two weeks, note hours saved, errors and human interventions needed, then decide.
  6. If the trial is not worth it, disconnect the apps, delete the dot and remove by hand the files and conversations it created.
Do not connect admin accounts, online banking or payment systems to a dot on trial. OpenAI hands password changes and money transfers back to the user; we suggest keeping them out of the trial in any case.
If the dot reads emails or documents holding customer or employee data, involve whoever handles privacy first. For Business workspaces OpenAI states that it does not use data to train its models by default, but admits that human review can occur in limited cases. The contract with the supplier, the legal basis and any impact assessment are still to be settled with your adviser.

What we do not know yet

  • What dots will cost after the first month: OpenAI will share usage terms for each plan later and has not named a price for additional dots.
  • When Pro users in the European Economic Area, Switzerland and the United Kingdom will get them: the exclusion applies at launch and OpenAI only speaks of reaching more users "soon", with no dates or countries.
  • Whether a Premium seat in ChatGPT Business is what OpenAI calls Business Premium: the dots pages use one name, the billing pages the other. Check in the workspace before paying.
  • Where a dot's data is processed: OpenAI's pages on dots do not say.
  • How they hold up in daily use: the protections are stated by OpenAI, and John Waldmann, CEO of Homebase, tells CNBC that millions of Muse downloads do not show whether people really use these agents.

Where we start

Dots are an OpenAI product that a company switches on from ChatGPT. The part of our work that this news touches is AI automation: language models integrated into the workflows already in use, with control over costs and answer quality. If a company asks us to assess an always-on agent, we propose to try it first inside a limited perimeter, starting from four things.

  • One task only, with a result that can be checked: as in every project, analysis comes before any switch-on.
  • A map of permissions: which apps the dot can read, which actions it can take alone and who approves the rest. Since its memory cannot be corrected entry by entry, whatever it must not see should not be connected.
  • A spending cap and an exit date, fixed beforehand, because the usage terms after the first month are not known yet.
  • A single page, readable by non-technical people, that sums up permissions, rules and who can stop the trial.

One limit: the choice of plan, the contract with OpenAI and the assessment of personal data stay with the company and its legal adviser; we handle the technical side of the trial.

If you want us to look at a task in your company and the perimeter of a trial, describe it through the contact page; the AI automation and chatbots service page explains how we work.

Sources

Share

Got a project in mind?

Let us build it together

We can help you turn an idea into a digital product that is solid, fast and ready to grow.

Related articles