Felix Laurent
OpenAI updated its ChatGPT dots on November 12, 2026. The always-on agents now reply faster, use less jargon and follow through on tasks better, according to OpenAI. They can also follow custom instructions, so users can set how a dot behaves and which actions need their approval.
The same evening, OpenAI's Thibault Sottiaux said a banked reset of usage limits had been activated and would reach accounts soon. It is the latest in a run of free resets OpenAI has handed out since autumn.
What changed in ChatGPT dots
OpenAI's dots are personal agents that live in the cloud, each with its own computer and browser. OpenAI launched them at DevDay on September 29. The November 12 update brings four changes, as OpenAI listed them:
- Faster replies.
- Less jargon in what a dot writes back.
- Better follow-through on tasks a dot takes on.
- Custom instructions that the dot keeps following over time.
OpenAI gave no benchmarks or numbers for the speed and follow-through gains. The custom instructions are the change users can actually see and test.
How custom instructions work in dots
There are two ways to set them. Users can tell the dot directly in a thread. OpenAI's examples are "always talk in lowercase" and "show me a draft before sending any emails" for email. Or they can write the rules in Customize > Behavior on mobile or in the dot's menu on web and desktop. The settings box holds up to 10,000 characters.
The instructions cover two kinds of things:
- how the dot behaves, such as tone, format or where it sends urgent updates;
- which actions need approval, such as emails, purchases or messages to other people.
OpenAI's documentation also describes custom rules, an optional set of firmer boundaries for ongoing work. Neither instructions nor rules can override the built-in safety requirements. The dot may also update its own instructions based on what users say in chats, according to the settings screen.
Those approval controls address a real complaint.
Why OpenAI needed this update
Users have been posting about dots on OpenAI's community forum for weeks. In the days before the update, one thread described repeated approval prompts and unexplained errors on tasks the user had already authorized. Another user wrote that a dot "feels less reliable than regular ChatGPT chat and codex" because of missed context and broken handoffs. Others reported voice calls that stall and answers that arrive as text instead.
"Better follow-through" targets the reliability side of that. Custom instructions target the approvals: a user can now tell the dot once which steps need a go-ahead, instead of answering a prompt each time.
The usage side got its own fix the same day.
What the banked reset is
A banked reset is a free, one-time refresh of Codex usage limits that OpenAI saves to a user's account. It does not take effect right away. The user chooses when to redeem it before it expires. On desktop it sits in Settings > Usage & billing, on the web in Usage settings, and in the Codex CLI under /usage.
| How a banked reset works | Details |
|---|---|
| What it restores | the main Codex allowance: 5-hour and weekly limits on Plus and Business Standard, the weekly limit on Pro |
| When it can be used | at least one covered allowance has less than 80% remaining |
| What it does not touch | image generation, file uploads and regular voice limits |
| Side effect | it can move the next scheduled refresh date |
| Cost | free, a discretionary promotion with no fixed schedule |
Sottiaux hinted at the reset on November 11. He wrote that Thursday was "a good time for a banked reset" in a post on X. He has handed out several before, including one in October and two on November 1. OpenAI has not said which plans get this one, though earlier banked resets went to paid accounts.
The resets matter for dots because a dot's chats are free, but its work is not.
Who can use dots and how usage counts
Dots are rolling out gradually to these plans:
- Pro 100, Pro 200 and Pro 500, for users 18 and older outside the European Economic Area, the UK and Switzerland;
- Business Premium;
- Enterprise, Edu and Healthcare, where access turns on by default from November 18, 2026, unless a workspace opts out.
For Pro and Business Premium, talking to a dot does not count toward ChatGPT usage limits. Tasks the dot starts in ChatGPT Work or Codex do count toward those products' limits. So a dot that runs a lot of Codex work burns the same allowance a banked reset restores.
Users can reach the same dot in ChatGPT, Slack, Teams or by phone call, and texting is in beta for Pro users in the US. OpenAI has also added GPT-6.1 Terra to the models that power dots.
The update continues a busy first six weeks.
How dots have changed since launch
Dots have changed almost every week since they launched.
| Date | Change |
|---|---|
| September 29 | dots launch at DevDay for Pro and Business Premium |
| October 9 | users can create dots on iOS and Android |
| November 9 | one dot can use several local computers |
| November 12 | faster replies, less jargon, better follow-through, custom instructions |
| November 18 | dots turn on by default for eligible Enterprise, Edu and Healthcare workspaces |
OpenAI is also not alone in personal agents. Meta's Muse, Google's Gemini agents and Anthropic's Claude Managed Agents compete for the same jobs. They book, shop, handle email and run work tasks in the background.
Is the dots update worth trying
Yes, if you already have a dot. Custom instructions are the most useful change. They let you decide once which actions need your approval and how the dot should talk to you. Setting "show me a draft before sending any emails" removes much of the risk of handing an agent your inbox.
The rest is harder to judge. OpenAI gave no numbers for faster replies or better follow-through, and the forum complaints about errors and reliability will need time to fade. The banked reset is a welcome bonus for heavy Codex users, but it is a one-time refresh, not a higher limit.
Sources
- ChatGPT's post on X – November 12, 2026
- Thibault Sottiaux's post on X – November 12, 2026
- Thibault Sottiaux's post on X about a banked reset – November 11, 2026
- Dots documentation, OpenAI
- How resets work in Codex, OpenAI Help Center
- Dot: repeated approvals and unexplained 403s, OpenAI Developer Community
- Dot continuity, access blockers and missed context, OpenAI Developer Community
- Dot voice calls connect but stall, OpenAI Developer Community
