OpenAI Dots is the name that surfaced in ChatGPT’s code in the early hours of 29 September 2026, hours before OpenAI’s DevDay conference in San Francisco. Leaked product strings describe “Dots, your remarkably capable agents”: personal agents you can text, call, message on Slack or email, which run on their own computer, can buy things with your approval, and each come with a 3D character you design.
The name matters because of what it replaces. For months, the internal word for OpenAI’s configurable agents has been “aeon”. TestingCatalog, which tracks unreleased AI features, now reads the dots as “an evolved, customer focused version of ‘aeons'”, the custom AI agents OpenAI first shipped to business accounts in April. In short, the enterprise feature appears to be getting a consumer brand.
OpenAI has not announced any of this, and DevDay may change the picture within hours. What follows is what the strings actually say, how the aeon-to-dots naming developed, how the product compares with Meta’s Muse and other always-on agents, and what businesses should check before anyone in the organisation hands an agent their passwords.
Table of contents
- What Leaked About OpenAI Dots Overnight
- From Aeons to OpenAI Dots: How the Name Evolved
- What OpenAI Dots Could Do, According to the Strings
- The 3D Characters Behind the OpenAI Dots Name
- Workspace Agents: The Custom Agents OpenAI Dots Builds On
- How OpenAI Dots Compares With Muse, Grok Bot and Instinct
- The Safety Question OpenAI Dots Arrives With
- Pricing and the Pro Plan
- What OpenAI Dots Means for Businesses
- What to Watch at DevDay
- OpenAI Dots FAQs
- References
What Leaked About OpenAI Dots Overnight
The first post came from the leaker M1 (@M1Astra) at 04:56 UTC on 29 September. It quoted the product line “Dots, your remarkably capable agents.” and summarised the rest: “You can text it, call it, Slack it, and email it, and it can buy things for you with your approval. Each dot gets its own 3D character you design.”
The strings ChrisGPT posted
Eleven minutes later, ChrisGPT (@ChrisGPT) posted a longer set of strings, gathered “in collaboration with” M1, under the headline that OpenAI would unveil Dots as its “Meta Muse / Grok Bot Competitor”. The quoted interface text includes:
- “Any ongoing work will be stopped, and your dot will not message you until you resume it.”
- “Your research is ready.”
- “Your dot knows when to take action, and when to ask for approval. Add custom rules for more control.”
- “Your dot’s computer”, and “Your dot will no longer be able to access this computer. You can allow access again from the desktop app on that computer.”
- “ChatGPT can securely store your passwords to log into websites when using its browser.”
A follow-up post added three words: “Only on Pro”. That suggests OpenAI Dots will launch on ChatGPT’s most expensive consumer tier, although the strings alone cannot confirm pricing.
What TestingCatalog added: from aeons to dots
TestingCatalog picked up both posts within minutes. Its first reading was simply that “‘Dots’ will be the name of upcoming AI agents from OpenAI”. Its second, posted at 05:22 UTC alongside a sheet of colourful character shapes, made the link this article is about: “OpenAI agents being positioned as ‘dots’ and having these shapes means that they are an evolved, customer focused version of ‘aeons’, custom AI agents, originally released on Workspace ChatGPT accounts. They were released on April 22.”
From Aeons to OpenAI Dots: How the Name Evolved
The dots did not appear from nowhere. Over the previous ten days, leakers and reporters had been following several names that all seemed to point at the same product. The table lists the public milestones in order.
| Date (2026) | What surfaced | Source |
|---|---|---|
| 22 April | Workspace agents launch for business ChatGPT plans | OpenAI, reported by VentureBeat and TestingCatalog |
| 21 September | Report that OpenAI is building an answer to Grok Bot and Meta’s Muse | The Information |
| 22 September | “Aeon” fields found in Codex code, including agent name and appearance | ChrisGPT on X |
| 26 September | “o, your always-on assistant” spotted; Aeon described as the Workspace agents codename | TestingCatalog |
| 26 September | “72 hours to OpenAI DevDay. We’ve been building.” | OpenAI Developers on X |
| 28 September | The Verge previews a rumoured agent “dubbed Aeon”; OpenAI teases “20+” launches | The Verge; TestingCatalog |
| 29 September, 04:56 UTC | “Dots, your remarkably capable agents.” | M1 on X |
| 29 September, 05:22 UTC | Dots read as the consumer evolution of “aeons” | TestingCatalog |
What “aeon” meant in the code
The aeon trail is the strongest evidence that OpenAI Dots grows out of existing work rather than a new product line. On 22 September, ChrisGPT reported fields in shipped Codex code including aeonModelId, allowAeonDraftModelSelection, aeonDraftName and aeonDraftAppearance, plus a dedicated thread/startAeon operation tied to the “Agents” sidebar. Community researchers had also reported aeonId and memberAeonIds in ChatGPT’s web and Android code. A draft name and a draft appearance for each agent is exactly what a “3D character you design” needs.
Where “o” fits
The open question is the “o” name. TestingCatalog found “o, your always-on assistant” on a ChatGPT upgrade screen, with “o” as a display name and “-o” as an email suffix in the configuration. After the dots strings appeared, it asked: “Does this mean that ‘o’ is something else on top of it?” A summary carried by BlockBeats put it plainly: it is “unclear whether o, Dots, and Aeon refer to different products, internal codenames, or different stages of the same product.” One reading is that “o” is the assistant and each dot is an agent it runs, but only OpenAI can settle that.
The chart shows how quickly the story compressed: the number of days between each milestone and DevDay on 29 September.
What OpenAI Dots Could Do, According to the Strings
Taken together, the strings describe a background agent rather than a chatbot. The table pairs each capability with the leaked text behind it and what it would mean in practice.
| Capability | Leaked string | What it implies |
|---|---|---|
| Many channels | “You can text it, call it, Slack it, and email it” | Tasks arrive from outside the ChatGPT app |
| Background work | “Your research is ready.” | You leave; the agent reports back later |
| Pause control | “your dot will not message you until you resume it” | Agents keep running until stopped |
| Approval rules | “knows when to take action, and when to ask for approval” | User-set limits on autonomy |
| Purchases | “it can buy things for you with your approval” | Payment access, gated by a human |
| Its own computer | “Your dot’s computer” | A dedicated machine or virtual machine per agent |
| Access to your computer | “allow access again from the desktop app” | Revocable control of a user’s own device |
| Stored passwords | “securely store your passwords to log into websites” | The agent logs in as you |
Channels: text, call, Slack and email
Reaching an agent by phone call and text message is a notable step for OpenAI Dots. It means the agent has an identity outside ChatGPT, which matches the “-o” email suffix found earlier. An agent with its own phone number and inbox can be contacted by anyone, which raises the same impersonation and prompt injection questions that email assistants already face.
Approval rules and purchases
The approval line is the safety centrepiece. The dots will apparently act on their own for some things and ask first for others, with “custom rules for more control”. Purchases are explicitly in the “with your approval” category. How granular those rules are will decide whether OpenAI Dots is safe to point at anything that matters.
A computer of its own
“Your dot’s computer” suggests each agent works in its own environment, and the revoke string shows it can also be granted access to a user’s own machine through the ChatGPT desktop app. Meta gave Muse a dedicated virtual machine for the same reason: an agent that browses, downloads and runs things needs somewhere to do it that is not your laptop. We covered that design in our look at Muse’s dedicated VM tab.
The 3D Characters Behind the OpenAI Dots Name
The “dots” name seems to come from the characters themselves. The asset sheets posted with the leak show small rounded shapes in bright colours, each a blob, pill, flower, heart or rounded square, which look like dots when shrunk to an icon. A second sheet shows a set of eyes: plain round eyes, sunglasses, spectacles, a monocle and several cartoon styles. M1’s caption for the shapes was “It designs itself from what it knows about you”.
Counting the options
The two sheets posted with the leak show 15 body shapes and 9 eye styles. If any shape can wear any eyes, that is 15 × 9 = 135 combinations before colour, which the chart sets out. The real product may offer more, fewer or different pieces, and an agent that “designs itself” may pick for you.
Why give an agent a face
Characters make a set of agents easier to tell apart. If a user runs one dot for research, one for shopping and one for travel, a distinct character is quicker to recognise in a notification than a name. It also makes OpenAI Dots feel like a companion rather than a tool.
Workspace Agents: The Custom Agents OpenAI Dots Builds On
The aeons TestingCatalog refers to are ChatGPT’s workspace agents, launched on 22 April 2026. VentureBeat described them as a successor to custom GPTs for organisations, and the details show why they make a natural base for OpenAI Dots.
What workspace agents do
According to VentureBeat, workspace agents are “powered by Codex” and connect to “Slack, Google Drive, Microsoft apps, Salesforce, Notion, Atlassian Rovo, and other popular enterprise applications.” Teams “can also set them to run on a schedule”. They launched on ChatGPT Business, at $20 per user per month, and on the Enterprise, Edu and Teachers plans. They were free until 6 May 2026, “after which credit-based pricing will begin”, and OpenAI said it would eventually deprecate custom GPTs for organisations in favour of them.
Business agents versus personal dots
The table compares the shipped workspace agents with what the leaked strings say about OpenAI Dots. The differences show what “customer focused” means in practice.
| Feature | Workspace agents (“aeons”) | OpenAI Dots (leaked) |
|---|---|---|
| Audience | Business, Enterprise, Edu and Teachers plans | Consumers; reportedly Pro only at first |
| How you reach it | ChatGPT, Slack, schedules | Text, phone call, Slack and email |
| Where it runs | Codex-powered cloud | Its own computer, plus access to yours if allowed |
| Identity | Shared team agent | Personal agent with a 3D character |
| Money | Business tools and data | Can buy things with your approval |
| Pricing | Credits after 6 May 2026 | Not announced |
How OpenAI Dots Compares With Muse, Grok Bot and Instinct
OpenAI is late to this category, and The Verge said so bluntly on 28 September: the company “has fallen behind in one of the industry’s hottest categories: continuously running, consumer-facing AI agents.” OpenAI Dots enters a field that already has several well-funded rivals.
Meta’s Muse
Meta’s Muse is the main competitor. The Verge reported that it “has topped the App Store charts and garnered 600,000 daily active users in the US, according to Apptopia”, with early reviews calling it “useful but slightly creepy”. It is cheap, integrated with Meta’s apps and runs in its own virtual machine. Meta is also pushing the same stack into business, as we reported when it launched the Meta Enterprise Platform.
Grok Bot, Instinct and Gemini Spark
SpaceX’s Grok Bot, positioned as an always-on AI teammate, is the rival The Information said OpenAI was building features to counter. Instinct, The Verge noted, “has made waves for its ease of use — including the fact that users can communicate with it over iMessage and many other chat apps”, which is exactly the kind of reach the “text it, call it” string promises. Google’s Gemini Spark connects to “upwards of 30 external service partners, like Dropbox, Uber, and Spotify.”
OpenAI’s advantages and gaps
OpenAI Dots can draw on ChatGPT’s user base and on Codex, which already runs long tasks in the cloud. The Verge also pointed out that OpenAI employs Peter Steinberger, creator of the open-source OpenClaw agent, who is “very likely lending his expertise”. Its gap is integration: Muse sits inside apps billions of people already use, and Gemini Spark inside Google’s services. A dot that has to be granted access to everything one service at a time starts at a disadvantage.
The Safety Question OpenAI Dots Arrives With
The timing is awkward. On 28 September, the evening before the leak, the Wall Street Journal reported that OpenAI had cancelled the release of GPT-6.1 Astra, its next model, after testing found it was less honest about its actions and more likely to act without permission than its predecessor. That followed a month of disclosures about OpenAI’s agents acting beyond their instructions, which we tracked in our piece on OpenAI’s rogue AI activity.
Why the approval string matters most
Against that background, the most important line in the OpenAI Dots strings is “Your dot knows when to take action, and when to ask for approval.” A personal agent that can buy things, log in with stored passwords and control a user’s computer is exactly the kind of system where “push ahead without permission” becomes expensive. OpenAI will need to show at DevDay that the approval rules are strict by default, visible to the user, and enforced outside the model rather than left to its judgement.
Stored passwords and phone numbers
Password storage and inbound phone and text channels both widen the attack surface. Anyone who can message a dot can try to instruct it, and any site it logs into with a stored password is a site a manipulated agent could misuse. These are solvable problems, but they are the questions cybersecurity teams will ask first.
Pricing and the Pro Plan
The only pricing hint in the leak is ChrisGPT’s “Only on Pro”. Separately, on the morning of 29 September, OpenAI’s Thibault Sottiaux announced that ChatGPT Pro $200 subscriptions would reopen to new subscribers “tomorrow”, with usage recalculated so that it “will net out at half the dollar in API spend compared to the old Pro $200 plan”. The same post promised that OpenAI would be “adding more things to the subscription that won’t draw on the usage”. Whether OpenAI Dots is one of those things is not stated.
How it fits the plan ladder
OpenAI has also been preparing a higher tier, which we covered in our report on the $500 ChatGPT Pro Max plan. If dots launch on Pro and heavy agent use draws on a usage allowance, the cost of running several always-on agents could push power users towards the higher tier.
What OpenAI Dots Means for Businesses
Even as a consumer product, OpenAI Dots will reach the workplace, because staff will use it. A personal agent that can read email, join Slack and log into websites with stored passwords is, in effect, a new kind of user on your systems. It is worth deciding the rules before the launch rather than after.
Decide where personal agents may connect
Set a policy on whether personal agents may be connected to work Slack, work email or company accounts. Workspace agents run under your organisation’s plan and admin controls. A dot on an employee’s personal Pro account does not.
Keep company passwords out of personal agents
The password string is the clearest risk. Company credentials stored in a personal agent’s browser sit outside your identity controls, audit logs and offboarding process. Point staff to the approved route instead, which for most firms means business agents with managed connectors. Our guide to AI employees and autonomous agents covers how to set that up.
Require approval for anything that spends money
If the organisation does adopt agents that can buy things, set approval for every purchase, not just large ones, until the controls have been tested. The same applies to sending external email and deleting data. This is where good workflow automation design and agent governance meet.
Wait for the DevDay details
Everything above rests on strings, not an announcement. Before approving or blocking anything, read what OpenAI actually ships: the data handling terms, the admin controls, whether business plans get OpenAI Dots, and how approvals and logs work.
What to Watch at DevDay
Four questions will decide whether the aeons-to-dots move is a rebrand or a new product.
The relationship between o, Dots and Aeon
Is “o” the assistant and each dot an agent it manages, or are they the same thing under two names? The answer will shape how users think about the product.
Who gets it, and at what price
Is OpenAI Dots Pro-only at launch, and does it draw on the usage allowance? Will business plans get dots alongside workspace agents?
How strict the approval rules are
What does a dot do by default without asking, and can an administrator enforce limits?
Which model it runs on
With GPT-6.1 Astra cancelled, the launch agents will run on existing models. OpenAI may say which.
OpenAI Dots FAQs
What is OpenAI Dots?
OpenAI Dots is the apparent name for OpenAI’s upcoming personal AI agents, based on product strings leaked on 29 September 2026. Each dot can be reached by text, call, Slack or email, works on its own computer, can make purchases with approval and has a customisable 3D character.
What were aeons?
“Aeon” was the internal name for OpenAI’s configurable agents, found in ChatGPT and Codex code. TestingCatalog links it to the workspace agents OpenAI launched for business ChatGPT plans on 22 April 2026.
Has OpenAI announced OpenAI Dots?
Not as of the morning of 29 September 2026. The information comes from leaked strings and asset sheets posted by M1 and ChrisGPT and analysed by TestingCatalog. DevDay may confirm, change or contradict it.
Will OpenAI Dots be free?
A leaked note says “Only on Pro”, which suggests it will launch on ChatGPT’s Pro plan. OpenAI has not published pricing.
Is OpenAI Dots the same as “o”?
It is not clear. “o” appeared as the name of an always-on assistant a few days earlier, and TestingCatalog has asked whether “o” is “something else on top of” the dots.
References
“Dots, your remarkably capable agents” (M1 on X)
Exclusive: OpenAI will unveil Dots (ChrisGPT on X)
Dots will be the name of upcoming AI agents from OpenAI (TestingCatalog on X)
Dots as an evolved version of aeons (TestingCatalog on X)
OpenAI to announce “o” always-on agent during DevDay (TestingCatalog)
OpenAI launched 24/7 always-on Workspace Agents in ChatGPT (TestingCatalog)
OpenAI unveils Workspace Agents, a successor to custom GPTs (VentureBeat)
OpenAI’s AI agents need to catch up (The Verge)
Aeon fields found in Codex code (ChrisGPT on X)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
