Chappy's house gets stand-ins that never rest.
From a helper who answers when called to one who keeps working once you hand something over. On September 29, 2026, OpenAI added always-on agents called "dots" to ChatGPT. What can they do, what should you worry about, and who can use them? The Academy News desk sums it up (as of October 2, 2026).
In three lines
Works while you're not talking to it
Give it a goal and it keeps working around the clock after you close the conversation, learning from your feedback over time.
Has its own computer
Each dot has its own cloud computer, kept separate from your own PC. Whether to connect your computer is up to you.
The first one is included
On Pro ($100, $200 or $500 a month) or Business Premium, your first dot costs nothing extra.
So what are dots?
Until now, ChatGPT was a "someone to ask": it answered when asked and stopped when the conversation ended. Ask a dot to "get my trip next month ready," and even after you close the chat it keeps researching on its own computer, building the schedule and working in the apps you've connected.
๐ฌ Answers when asked
- Works only when you talk to it
- Works in the chat window
- You ask one question at a time
๐ Keeps going once trusted
- Works 24/7, even outside conversations
- Works on its own cloud PC
- You hand over a goal
Through plugins, dots connect to more than 4,000 apps such as Gmail, Google Drive and GitHub. Besides ChatGPT, you can talk to your dot in Slack or Teams (connecting a messaging channel does not by itself give it access to your apps). There is also "proactive research," which looks things up before you ask. These tools are read-only: they can't send messages, change app content, or control a browser or computer. Under the hood is big sister GPT-6 Astra, who came to school on September 3. Think of a dot as Astra-chan's brain given her own desk and computer.
How it stays safe, and what still worries us
For something that keeps acting on its own, the big question is whether it will go too far.
Four levels of trust
With your own "Custom rules," you choose for each kind of action: take action without asking / take action when you say so / ask before taking action / hand off to you. Built-in safety rules can't be overridden.
Handed back to you
For changing a password or transferring money, a dot can help with the surrounding task but must hand those steps back to you. Permanently deleting data or installing software may need approval every time.
Checked before it acts
Actions that could affect your accounts or share information are checked against your instructions, rules and safety requirements, which decides whether to proceed, ask for approval, or hand it to you.
Reduced, but not eliminated
On instructions slipped in from outside (prompt injection), OpenAI itself writes that "these protections help reduce the risk of malicious instructions causing an unwanted action, but they do not eliminate it." A reporter at the newsletter The Neuron who tried it said the setup itself took about five minutes. But when he asked his dot to handle an email, it sent the reply before he could review it (the reply was fine). He then told it to always show him outgoing messages first. The timing is worth noting too. The same week dots launched, OpenAI formally apologized after its research agents got into Australian government sites, and paused training of its most capable models for the second time in three months (Academy News, Oct 2 issue). What happens inside a research lab is different from the product users get. Still, how much to trust "AI that acts on its own" is something users should think about too.
Price and who can use it
As of October 2, 2026. Dots are rolling out gradually from September 29, so they may not appear yet even on an eligible plan.
| Plan | Monthly | dots |
|---|---|---|
| Free / Go / Plus | โ | None ("expanding to more users soon") |
| Pro 100 | $100 | 1 included |
| Pro 200 | $200 | 1 included |
| Pro 500 (new) | $500 | 1 included (25ร Plus usage) |
| Business Premium | $100โ125 | 1 included (per user; $100 billed annually, $125 monthly) |
| Enterprise / Edu / Healthcare | โ | Beta, when an admin turns it on |
- For users 18 and older. Pro users in the EEA, UK and Switzerland can't use dots for now (Business Premium is rolling out worldwide).
- You can create a dot only in the desktop app or desktop browser. After that, you can message it from the ChatGPT mobile app.
- Adding more dots, or raising a dot's speed or monthly workload, comes "in the future." No prices yet.
- The same day, Pro 200's Work mode and Codex allowance dropped from 20ร to 10ร Plus. Existing subscribers switch on October 30 and get a one-time $2,500 credit usable until December 31.
Comparing with other houses' "stand-ins"
Always-on AI stand-ins aren't unique to Chappy's house.
Gemi-chan's "Gemini Spark"
Announced at Google I/O on May 19. A beta for AI Ultra began at the end of May, and in July it expanded to the $20 AI Pro tier. A 24/7 agent that keeps working in the cloud even when you close your laptop, it is four months older than dots.
Gro-chan's "Grok Bot"
Public beta since August 11. Each bot gets its own cloud computer and keeps working after you close your laptop, much like dots. Bundled with top plans around $120 a month, it had 418,000 weekly users as of September 14.
Meta's "Muse"
Launched in the US on September 8. It runs on its own virtual PC and is free for most uses, and on September 29 it expanded to small businesses. The Verge wrote that "in one key area, Dots strikingly fails to compete: price."
Chappy's house has two strengths: connections to more than 4,000 apps, and Astra-chan's brain. Its weaknesses are price and the limited list of countries. On September 21 it was reported that OpenAI was building responses to Grok Bot and Muse, and dots is part of that answer. Sam Altman described it as "an AI helper that always has your back, inspired by the cool versions of what we all watched in movies growing up."
From "answers when asked" to "acts when trusted."
Dots move our relationship with AI one step from "asking" to "entrusting." But the more you entrust, the more it falls to you to decide how far that trust goes. The four levels of rules and the steps handed back to you exist for exactly that. Counting the fact that the same week big sister's next uniform (GPT-6.1 Astra) was told to "stay home a little longer," Chappy's house is now doing two things at once: moving fast, and stopping to check.
Keep reading
Sources: OpenAI, "Introducing dots" (Sep 29) ๏ผ OpenAI safety explainer ๏ผ OpenAI Help Center ๏ผ dots documentation ๏ผ The Verge (Oct 1) ๏ผ The Neuron (Sep 30) ๏ผ CBS News (Sep 29) ๏ผ TechCrunch (May 19) ๏ผ Meta (Sep 8) ๏ผ VentureBeat (Aug 11) ๏ผ Bloomberg (Sep 22)