Most setup guides for a new AI agent start with the fun part: connect everything and see what happens. For an agent that can run in the background and use your accounts, that order is backwards. This guide shows how to set up OpenAI dots securely by locking things down first and opening access one step at a time. It covers where your dot works, which apps to connect, how to write rules before the first task, how to handle channels like Slack and texting, and how to test before you trust. For hands-on grounding in the programming and tooling behind agents, the Tech Certifications from Global Tech Council are a solid starting point.

What are OpenAI dots, and why does setup order matter?

A dot is an always-on agent inside ChatGPT, powered by GPT-6 Astra, with its own cloud computer. It works across the apps you connect, remembers context and keeps going between conversations. OpenAI launched dots on September 29, 2026.

Setup order matters because a dot’s reach grows with every choice you make. Connect an app and it can read from it. Allow local access and it can work on your files. Add a channel and other people can see what it posts. Each step is easy to undo in theory, but information a dot has already gathered doesn’t always go away when you reverse the step. Starting narrow means you decide what to add, instead of cleaning up what you added too soon.

Decide where your dot works

Every dot runs on its own cloud computer, and that is the safest place to begin. Access to your local computer is optional and starts off. Your dot’s cloud computer also doesn’t automatically inherit your local VPN, browser sign-ins or device policies, which keeps your day-to-day environment separate from the agent’s. The flip side is that a site may require you to sign in on the cloud computer before your dot can use it.

If you later need local access, enable it deliberately. In the ChatGPT desktop app on the computer you want to use, open the dot’s profile, find Your computer under Computers and select Allow access. Work on your own machine runs in separate tasks that appear in the sidebar. With local access on, a dot can work on files there, use local skills and fall back to your local browser when its cloud browser is blocked. The computer has to stay online with the app open, and you can revoke access whenever you like.

Many people never need local access for the first few weeks. Treat it as an upgrade you earn through good behavior, not a default.

Connect fewer apps than you think you need

Apps give a dot information it can use for your tasks. To connect one on mobile, open the dot’s profile, go to Customize and then Plugins. Select a supported app and read its permissions first. Connect only what your first task needs.

Two details deserve attention. First, connected apps let your dot review information proactively and form memories from it, even when you haven’t asked a specific question. That’s part of what makes it useful, and it’s also why a broad connection is a bigger decision than it looks. Second, disconnecting an app doesn’t delete information the dot has already obtained. To remove that, you need to delete the dot.

Logins are handled with care: OpenAI says dots use saved passwords without exposing them to the model. You can also connect a personal email account for your dot to use, though it can’t have its own address at launch. Give email access only when a task truly needs it.

Write your rules before your first task

Custom rules are where you decide what your dot does alone. In the dot’s settings, open Custom rules to review the defaults, then add rules by describing an action and choosing one of the behaviors: take action without asking, take action if pre-approved, ask before taking action, or hand off to you. OpenAI notes that “pre-approved” means you explicitly requested the action in your prompt.

You can set rules for what your dot may share, purchase or access. Good starter rules are narrow and cautious:

Ask before sending anything outside your own notes.
Ask before any purchase, however small.
Hand off anything involving account changes or payments.
Allow reading and summarizing without asking, for the apps you connected.

Marketing teams can apply the same idea. A dot that drafts campaign emails can prepare them and hand off, while a person presses send. Teams that want to connect that kind of automation to real campaign work can pair it with a Digital Marketing Course so marketers know which actions deserve sign-off. Remember too that your rules sit on top of built-in safeguards, and OpenAI says your dot can still make mistakes when following them.

Secure the channels

Channels turn a private helper into something other people can see.

In Slack, a dot joins supported workspaces and posts with its own identity. Only the dot’s owner can direct it, so other people’s direct messages or mentions don’t start work. But when it posts in a channel, everyone in that channel can read the post. Don’t ask it to summarize private material in a shared space. If your Slack workspace requires app approval, a workspace owner or app manager may need to approve the installation.

Texting is a limited beta for Pro users in the US, it runs through a third-party provider, and OpenAI advises caution with sensitive information. Message and data rates may apply. You can reply STOP to stop outgoing texts. For most security-minded setups, leave texting off until you have a reason to use it.

Readers who work across AI, security and Web3 systems can find wider context in the Deep Tech Certifications from Blockchain Council.

Test with a harmless task, then review

Before handing over real work, run something low-risk, such as reviewing your calendar and summarizing the week. Open the dot’s profile to see activity under In progress, Scheduled and Completed, and open its computer to watch what it does. Compare the output with what you expected, and check any detail that matters before relying on it.

Remember that OpenAI describes dots as capable of mistakes, so consequential work should be reviewed. Add one app, one rule or one scheduled task at a time and look at the result each time. If something surprises you, tighten the rule that should have caught it before you add anything else.

Know your off switches

Four controls behave differently, and it helps to know what each one leaves behind.

Pause stops the dot until you resume it, and keeps its data.
Revoke access on your computer stops local file access and work on that machine.
Disconnecting an app stops future access, but the dot keeps what it already learned.
Reset deletes the dot, its conversations, saved memories and scheduled tasks. Files, Codex threads and ChatGPT conversations it created are stored separately.

Conclusion

To set up OpenAI dots securely, work from the narrowest position outward. Start in the cloud and leave local access off. Connect only the apps your first task needs, write cautious custom rules before you assign anything, and keep Slack and texting out of the picture until they earn a place. Run a harmless test, review the output and widen access one step at a time. Know what Pause, Revoke access, disconnecting an app and Reset each do, because none of them is a full undo. The safest dot is the one whose access you can explain in a sentence, and every addition should still pass that test.

FAQs

1. What does setting up OpenAI dots securely mean?

Starting with minimal access and adding apps, computers and channels one step at a time.

2. Is local computer access on by default?

No. It’s off, and you enable it in the ChatGPT desktop app.

3. Why start with the cloud computer?

It keeps your own files and device out of reach, and it doesn’t inherit your VPN or device policies.

4. Why might a website need a sign-in on the cloud computer?

The cloud computer doesn’t automatically carry your browser sign-ins.

5. How do I turn on local access?

Open the dot’s profile in the desktop app, find Your computer under Computers and select Allow access.

6. How do I revoke local access?

Use Revoke access in the same place to stop file access and work on that computer.

7. How should I choose which apps to connect?

Connect only what your first task needs and read each app’s permissions first.

8. Do connected apps give a dot background access?

Yes. A dot can review connected information proactively and form memories from it.

9. Does disconnecting an app delete what the dot learned?

No. To delete that information, you need to delete the dot.

10. Does the model see my saved passwords?

OpenAI says dots use saved passwords without exposing them to the model.

11. What are custom rules?

Rules that set how your dot handles specific actions, such as sharing, purchasing or accessing.

12. What does pre-approved mean?

You explicitly requested that action in your prompt.

13. What starter rules make sense?

Ask before sending outside your notes, ask before purchases and hand off account or payment changes.

14. Who can direct my dot in Slack?

Only its owner. Others’ messages or mentions don’t start work.

15. Can others see my dot’s Slack posts?

Yes. When it posts in a channel, other members of that channel can see it.

16. Is texting safe for sensitive details?

OpenAI advises caution because it uses a third-party provider.

17. How do I stop texts?

Reply STOP to stop outgoing text messages.

18. How should I test a new dot?

Use a harmless task, watch its work in the dot’s computer and verify details before relying on them.

19. What’s the difference between Pause and Reset?

Pause stops the dot and keeps its data. Reset deletes the dot, its conversations, memories and scheduled tasks.

20. Can a dot still make mistakes with good rules?

Yes. OpenAI says so, which is why review matters.

How to set up OpenAI dots securely: a security-first configuration guide