# OpenAI says teams of dots will come "over time." A viral setup guide already has you hiring five dots

**OpenAI dots, as of October 6, 2026, give each user one primary dot that runs on GPT-6 Astra and cannot hand work to another dot.** OpenAI's September 29, 2026 [launch announcement](https://openai.com/index/introducing-dots/) is explicit that today you start with your primary dot. Teams of dots are something OpenAI says it envisions "over time." A widely shared setup guide from the days after DevDay skips that line. It has readers hire five dots, give each one a different model, and let a chief-of-staff dot route tasks to the other four.

The guide is not useless. Several of its operating details match OpenAI's documentation almost word for word. Its architecture is the problem. Below, each claim is checked against the docs, and the parts that hold up are rebuilt as a setup for the single dot you actually get.

#### The blueprint gets the off switch right and the org chart wrong

As of October 6, 2026, OpenAI's own documentation is the only source for how dots behave. The claim-by-claim check of the viral five-dot blueprint below is ours.

| Blueprint claim | What OpenAI documents (October 6, 2026) | Verdict |
|---|---|---|
| One account can staff a team of several dots | You start with your primary dot; more dots come later | Not yet |
| A dot can switch between GPT-6 Astra and GPT-6.1 Sol | Dots are powered by GPT-6 Astra; Enterprise model controls and defaults do not apply | No |
| A chief-of-staff dot delegates subtasks to worker dots | A dot delegates to background agents, cloud threads, and Work or Codex tasks | Partly: the split exists, the workers are not dots |
| A Role Description field holds each dot's standing contract | No such field; preferences go in conversation, action limits in Custom Rules | Undocumented |
| A site stays permanently signed in | A session lasts until you sign out or the site expires it | Overstated |
| Dots react to webhooks | Dots can respond to specified events where a connected service supports event monitoring; webhooks are never mentioned | Webhooks undocumented |
| Moving background work to Sol cuts spend by more than 70% | Sol's API token price is one-fifth of Astra's, but dots have no model setting | Math checks out, no lever |
| Dots keep working with your laptop closed | Cloud work can continue while your devices are off | Correct |
| Pausing the main agent does not cancel scheduled jobs | Pause stops the current main task; it does not stop every delegated task or cancel future runs | Correct |
| Adding a dot to Slack does not start monitoring | Adding a dot to a channel does not establish a monitoring schedule | Correct |

Every wrong row comes from one assumption: that dots are already a platform for composing agents. What OpenAI shipped is a personal agent with its own cloud computer that can farm out work.

#### One dot can still split work across agents and Codex

OpenAI's getting-started guide says you will be able to add more dots in the future. The launch post also previews specialist dots, which get their own identity for access management, IT-provisioned hardware and deep integrations with a company's systems of record. Those start as focused enterprise pilots, and OpenAI is working with Microsoft to manage them through Agent 365.

So Vance, Mara, Cole and the rest of the five-person roster can't each become a dot today. The problem the roster was meant to solve does have an answer. The [tasks and memory documentation](https://learn.chatgpt.com/docs/dots/tasks-and-memory) lists four kinds of work a dot can hand off:

| Work type | What the dot does with it | Requirement |
|---|---|---|
| Background agents | Runs several things in parallel; agents report back to the dot | None |
| New cloud threads | Opens separate conversations you can steer in desktop, web and mobile apps | None |
| New local Work or Codex task | Creates the task on a computer connected to the dot | Computer online with the ChatGPT app open |
| New cloud coding task | Creates the task in a Codex cloud environment | Set up the environment and repository in Codex first; your computer can be offline |

![Diagram: the blueprint wants five dots, while one primary dot hands work to background agents, cloud threads, local Work or Codex tasks and Codex cloud tasks](https://s4.tenten.co/learning/content/images/2026/10/english-infographic-1.png)

Each new task gets only the instructions and context the dot passes along. It does not inherit every conversation you have had with the dot. That is the context isolation the blueprint describes, happening inside one dot. OpenAI's privacy and safety FAQ adds that action rules and approvals carry over into delegated work.

My read: the roster's real value was dividing responsibilities, and one dot can already hold more than one ongoing responsibility, per OpenAI's own guide. What it can't give you is a separate identity and permission set per role. That waits for specialist dots or additional dots.

#### A 70% saving from GPT-6.1 Sol has no switch inside dots

The [GPT-6.1 Sol model page](https://developers.openai.com/api/docs/models/gpt-6.1-sol) lists $2 per million input tokens, $10 per million output tokens and $0.10 for cached input. GPT-6 Astra's model page lists $10, $50 and $1.00. That is exactly one-fifth, or 80% cheaper, by our arithmetic on the published prices.

Both models share the same API limits: a 1,050,000-token context window, 922,000 maximum input tokens and 128,000 maximum output tokens. The blueprint says Sol can swallow 1.05 million tokens in one prompt. That figure is the total context window, and Astra's is identical.

None of this applies inside dots. OpenAI's [workspace admin article](https://help.openai.com/en/articles/20001554-manage-dots-in-chatgpt-workspaces) says Enterprise model controls and default model settings do not apply to dots, and no dots page documents a model picker. Sol exists elsewhere in ChatGPT. OpenAI's September 29, 2026 Sol announcement made it available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users, but not yet in Chat.

Dots aren't billed per token either. According to the launch post, conversations with your dot don't count toward ChatGPT usage limits. Tasks the dot starts in Work or Codex count toward those products' limits as usual. Your plan includes an allowance for deeper work, with extended limits in the first month. OpenAI doesn't say which model a dot-created Codex task uses, so don't build a budget around it.

The same goes for GPT-6 Astra Ultrafast. OpenAI's DevDay 2026 recap puts it at up to 8x faster token generation in Codex, or 300 tokens per second, on Pro 500 and Enterprise plans. It isn't a dots setting, and 8x describes generation speed, not an 8x shorter development cycle.

#### Set up one dot in four moves

OpenAI's own advice is to start with work you check or update regularly, describe the responsibility, share sources, and review the first result before widening scope. The four steps below are our recommended order, built from the documentation. OpenAI doesn't prescribe it.

##### Send the role brief as your first message

There is no Role Description field. You can set a name, appearance and handle, and the dot keeps its own notes on your preferences. The blueprint's role card still works once you drop the Model field, which doesn't exist. Send it as a message:

```text
You are the centralized Operations Coordinator for product releases.
Your sole responsibility is auditing launch readiness across verified briefs,
team discussions, and project checklists.

Operational Rules:
1. Always link every factual claim directly to an accessible source document.
2. If two sources contradict each other, highlight the conflict explicitly.
3. Never send external messages, modify shared production files, or assign
   binding tasks to colleagues without explicit human approval.
```

##### Turn each approval line into a Custom Rule

The viral blueprint's permission block is a fine conversation instruction:

```text
Permission Guidelines:
1. Autonomously read, analyze, and prepare drafts across all connected services.
2. Always halt and request explicit approval before:
   * Sending an email or messaging a client.
   * Modifying or deleting shared database entries.
   * Triggering financial transactions or purchasing credits.
   * Publishing code or modifying production branches.
```

Instructions in chat get interpreted case by case. For a standing boundary, make each bullet a rule. The [Control your dot page](https://learn.chatgpt.com/docs/dots/controls) puts it under Settings > Personalization > Custom rules, in the Permissions section. Select Add, describe the action, pick a behavior, then select Add rule. There are four behaviors:

| Rule | What it does | Where the blueprint's actions fit |
|---|---|---|
| Take action without asking | Acts without approval | Reading, analysis and drafting usually need no rule |
| Take action when you say so | Acts only when you explicitly request it | A post or update you name in your prompt |
| Ask before taking action | Asks for approval every time | Messaging clients, editing a shared database |
| Hand off to you | Asks you to do it yourself | Deleting shared files, pushing to production branches |

OpenAI's Help Center calls the second option "Take action if pre-approved," meaning you explicitly requested the action in your prompt. Only two placements in the right-hand column come from OpenAI: "Ask before taking action" for messages to customers and "Hand off to you" for deleting shared project files. The rest are our suggestions.

Rules don't grant access to any app or computer; app actions live separately under Plugin permissions. Rules also can't override built-in safety requirements. Changing a password or moving money between financial accounts is always handed back to you. Permanent deletion, installing software from an unrecognized source and granting new security-sensitive access need your confirmation each time. If a workspace admin turns custom rules off, saved rules stop applying.

##### Connect plugins and channels separately, then sign in

Apps connect through plugins, which OpenAI says reach more than 4,000 apps. Those permissions are shared across ChatGPT, ChatGPT Work, Codex and dots, and you can let a dot read email without letting it send. Slack and Microsoft Teams are messaging channels you add from the dot's profile. Adding a channel doesn't open your inbox, other apps or your computer.

When a site needs a login, the dot sends a private sign-in form, and your credentials go to the cloud browser outside the conversation. You can also select Take over, sign in yourself, then select Return control. Per OpenAI's [computers and apps guide](https://learn.chatgpt.com/docs/dots/computers-and-apps), the session lasts until you sign out or the site expires it. The cloud browser doesn't inherit logins from your own browser, and some sites block cloud browsers outright.

##### Give every schedule a time zone, end date and destination

The viral blueprint's morning brief works as written:

```text
Schedule this routine:

Every weekday at 08:30 Europe/London, execute the Morning Market & Pipeline Audit.

Use the verified sources in our Space page and the connected CRM dashboard.

Compile a briefing with four sections:
1. Critical Account Changes (last 24 hours).
2. Blocked Issues & Missing Owners.
3. Competitor Announcements.
4. Decisions Waiting on Me.

Deliver the final brief directly to the #leadership-briefing channel in Slack.
Confirm the timezone, schedule, and destination.
```

OpenAI asks for four things in a recurring task: what to check, when to run including time zone and end date, which changes deserve a notification, and where results go. This prompt has no end date; OpenAI's own example runs "for the next four weeks." Posting to a team channel adds a check, because sharing private details with other people needs your permission. Afterward, open Scheduled and confirm the instructions, timing and destination.

Beyond fixed times, a dot can respond to a specified event when the connected service supports event monitoring. OpenAI's example is a new bug report in a Slack channel. Ask the dot to confirm which events it can follow, since connecting Slack alone creates no monitoring task.

#### Each of the four blueprints runs partly on one dot today

| Blueprint | What one dot can do today | What changes or waits |
|---|---|---|
| Launch coordinator: 08:30 daily brief, triggered by new Linear blockers | Scheduled brief, sourced summary, follow-up drafts held for approval | A Linear trigger depends on whether that service supports event monitoring |
| Lead intelligence desk: Sol scanning around the clock | Cloud-browser research behind your login, a staging table, outreach drafts never sent | No Sol; replace "continuous" with a schedule or let the dot decide when to follow up |
| Dependency and security auditor: Sol plus Codex | Tasks in a Codex cloud environment you built, returning patch PRs for review | Per-PR triggers and a dots link to Codex Security Cloud are undocumented |
| Meeting follow-up desk: triggered when a call ends | Decisions and action items from notes you share, follow-ups held for approval | No documented end-of-call trigger; the Meetings plugin is a macOS desktop beta |

All four of the viral blueprint's agent designs run partly on one dot today. What survives runs on schedules, drafts and approvals. What breaks is either a model choice or a trigger OpenAI hasn't documented.

#### Stopping a dot takes three separate actions

The viral dots blueprint is right about one thing: Pause alone doesn't stop everything. Per the Control your dot page, Pause stops the dot's current main task. It doesn't stop every delegated task or cancel future scheduled runs. Stop delegated tasks one by one in Activity, disable or delete recurring work in Scheduled, and use Resume to restart.

![Diagram: Pause stops the main task, Activity stops delegated tasks, and Scheduled disables recurring runs](https://s4.tenten.co/learning/content/images/2026/10/english-infographic-2.png)

Stopping work doesn't undo completed actions. Deleting a dot doesn't recall delivered messages or reverse changes already made in connected apps. OpenAI's safety write-up adds that monitoring can pause a dot's work and show you a warning if it flags behavior such as acting outside your instructions.

#### Frequently asked questions

##### Can I create a second OpenAI dot?

OpenAI dots don't support a second dot as of October 6, 2026; each user starts with one primary dot. OpenAI's Help Center says more dots are coming, and the launch post calls teams of dots a longer-term vision. Specialist dots for companies are limited to enterprise pilots.

##### Can an OpenAI dot run on GPT-6.1 Sol instead of GPT-6 Astra?

OpenAI dots can't switch models. Dots run on GPT-6 Astra, no dots page documents a model picker, and Enterprise model settings don't apply. GPT-6.1 Sol is available through the API and in ChatGPT Work and Codex.

##### Which ChatGPT plans include OpenAI dots in the US?

OpenAI dots are rolling out to Pro 100, Pro 200 and Pro 500 users over 18 outside the European Economic Area, the UK and Switzerland, so US Pro subscribers qualify. Business Premium is rolling out worldwide. Enterprise workspaces have dots off by default until an admin enables them.

##### Can an OpenAI dot react to new GitHub issues or Linear tickets?

OpenAI dots respond to specified events only when the connected service supports event monitoring, and OpenAI's example is a new bug report in Slack. OpenAI doesn't list whether GitHub or Linear qualify. The word webhook doesn't appear in the dots documentation.

#### Sources

- [OpenAI: Introducing dots (September 29, 2026)](https://openai.com/index/introducing-dots/)
- [OpenAI: How we build safety, security, and privacy into dots](https://openai.com/index/how-we-build-safety-security-and-privacy-into-dots/)
- [OpenAI Help Center: Getting started with your dot](https://help.openai.com/en/articles/20001530-getting-started-with-your-dot)
- [OpenAI Help Center: Dots privacy, security, and safety FAQs](https://help.openai.com/en/articles/20001529-dots-privacy-security-and-safety-faqs)
- [OpenAI Help Center: Manage dots in ChatGPT workspaces](https://help.openai.com/en/articles/20001554-manage-dots-in-chatgpt-workspaces)
- [ChatGPT Learn: Tasks and memory](https://learn.chatgpt.com/docs/dots/tasks-and-memory)
- [ChatGPT Learn: Control your dot](https://learn.chatgpt.com/docs/dots/controls)
- [ChatGPT Learn: Dots](https://learn.chatgpt.com/docs/dots)
- [ChatGPT Learn: Channels](https://learn.chatgpt.com/docs/dots/channels)
- [ChatGPT Learn: Connect computers and apps to your dot](https://learn.chatgpt.com/docs/dots/computers-and-apps)
- [OpenAI: Introducing GPT-6.1 Sol](https://openai.com/index/introducing-gpt-6-1-sol/)
- [OpenAI API: GPT-6.1 Sol model page](https://developers.openai.com/api/docs/models/gpt-6.1-sol)
- [OpenAI API: GPT-6 Astra model page](https://developers.openai.com/api/docs/models/gpt-6-astra)
- [OpenAI: DevDay 2026 Recap](https://openai.com/index/devday-2026-recap/)

#### Author Insight

The case for waiting on more dots is about permissions, not workload. Custom Rules apply to the whole dot, and delegated work inherits them. If your blueprint needs two different permission sets, say one role that edits the CRM and another that only reads code, wait for additional dots or specialist dots. If every job can live inside one approval boundary, one dot plus Codex cloud tasks covers it today.