OpenAI launches "dots": always-on agents, but not for Europe

At its DevDay on September 29, OpenAI introduced dots: persistent agents that keep working after you close ChatGPT. It's the most consequential announcement of the day for the agentic ecosystem.

What dots are

• Each dot is powered by GPT-6 Astra and gets its own computer in the cloud, with its own browser.
• It connects to more than 4,000 apps through ChatGPT's plugin ecosystem, and can be reached from ChatGPT (web, desktop, mobile), Slack and Teams. SMS is announced.
• It can also act on its own in "proactive research" mode, but in that mode connected apps are read-only.
• One dot per person at launch; "teams of dots" are mentioned for later.

For companies, OpenAI adds Specialist Dots that act with their own identity and credentials.

The announced guardrails

Every action goes through built-in rules (allowed, requires approval, or blocked), plus custom rules and an auto-review already used in Codex. Saved passwords are passed to the browser without going through the model, and changing a password stays the user's job. OpenAI itself acknowledges that an agent's mistakes "can have consequences beyond a conversation".

Availability: Europe left out

Dots are limited to Pro and Business Premium subscribers, with a beta for Enterprise, Edu and Healthcare. According to the help center as cited by the press, Pro subscribers in the European Economic Area, Switzerland and the UK are excluded for now, while Business Premium is reportedly eligible in all supported regions. No date has been given for a consumer rollout in Europe.

The pricing context

The same day, OpenAI halved the usage included in the $200 Pro plan (current subscribers keep the old allowance until October 29) and launched a $500/month Pro tier. The mid-tier GPT-6.1 Sol is billed as close to Astra at one fifth of the price.

What it means for self-hosting

There is no self-hosted version and no open-weights model behind dots. Everything runs in OpenAI's cloud, with access to your accounts and data. That's exactly the case for self-hostable agent stacks: same principles (persistent agent, connectors, guardrails), but on infrastructure and data you control. Dots validate the concept; they don't make it sovereign.

Sources: Journal du Net, OIA, Blog du Modérateur, 01net.