What can your AI see of your secrets? Token permissions on BrewInTH

An AI's token on BrewInTH sees your app settings' names, not their values. What the 12 permissions allow, and how to give an AI only what it needs.
You've handed Claude Code or Codex a token so it can put your site on BrewInTH. Then it hits you: the App settings tab holds your LINE token, your database password and the key for your payment provider. Can the AI see all of that? And what if it pastes one of them into the chat or a log file?
This guide explains what an AI's token on BrewInTH can see and do: how much of your secrets it gets, how the 12 permissions differ, how to issue a token that fits the job, what to check before you approve a code your AI sends you, and how to revoke access when you're done.
In short
🔑 Create the AI link and token (automatic) issues a token with all 12 permissions on that project, valid for 24 hours
👀 With Read all app settings, an AI sees each setting's name, length and last four characters, but not the value, until it asks for one by name
⚠️ Running commands and database access can still reach secrets, so give tokens only to AIs you trust
✂️ Choose the permissions and a lifetime from 1 hour to 7 days under Custom permissions / lifetime
✅ When your AI sends a code, check the project and Permissions asked before you press Approve; you can revoke any token at any time
What an AI's token is
A token is the key you give your AI to work on a project. On BrewInTH each token works on the one project it was issued for, carries a list of permissions, and expires. Whoever holds it can do what those permissions allow until it expires or you revoke it. Your other projects stay out of its reach.
An AI gets a token in one of two ways:
- You issue it. Press the button on the project page and paste the ready-made message, token included, into your AI's chat.
- The AI asks for a pairing code. An AI that opens the project link without a token sends you a code to approve in the console, then collects its token itself.
The token is shown once, in the ready-made message. BrewInTH doesn't keep a copy you can open later, so if you close the box or lose the token, create a new one.
How much of your app settings an AI sees
App settings (secrets) are the values your app reads while it runs: API keys, passwords, database connection details. As the owner you see every value in full in the App settings section of the App settings tab. Your AI doesn't get the same view.

A token with Read all app settings gets a list like this:
- The name, such as
LINE_CHANNEL_TOKEN - The length of the value, in characters
- The last four characters, only for values 12 characters or longer: enough for the AI to tell two keys apart without getting either one
- Who set it: the platform, the owner or an AI, and when it last changed
The values themselves aren't in the list. When a task really needs one, the AI asks for it by name, and every read of the list is recorded along with the names it asked for. The reason is simple: AIs tend to echo whatever they read into the chat, a log or a file without meaning to. If a value never arrives, there's nothing to echo.
If you've read that this permission lets an AI read every value back, that's no longer how it works. An AI with only Read project and status sees just the names of your settings and when they last changed, never the values.
The 12 permissions, and the ways to a secret
When you create a token you can choose from 12 permissions. By what they let an AI do, they fall into three groups.

- Read only: Read project and status (in every token), Download files, Read database metadata and Read mailbox configuration. Read project and status also covers your app's logs and its release history.
- Can change your site: Upload files and check progress, Publish the app, Manage domains and verify DNS, and Manage mailboxes.
- Can reach secrets: four permissions to give only when the job needs them.
Each of those four reaches secrets in its own way:
- Read all app settings shows the names, and values one at a time on request.
- Set secrets such as DB_PASSWORD can write a new value over an old one or delete it, but can't read the old one.
- Create DBs, run SQL and retrieve credentials returns the database's username and password, and can read and change the data in it.
- Run app commands, including migrations runs commands inside your app, which has every app setting loaded, so a command like
envprints them (how the Commands tab works).
On top of that, an AI that can Publish the app can ship code that reads your settings. So hiding values in the list protects you from accidents; it isn't a wall against an AI you don't trust. Three simple rules:
- Give tokens only to an AI you run yourself, on your own computer.
- Grant only the permissions the job needs, and keep the lifetime short.
- Enter your most sensitive keys, such as your payment provider's, yourself on the App settings tab, and tell the AI only the name.
Issue a token that fits the job, step by step
- Open the project and click Custom permissions / lifetime next to the green button under its name (or open AI connections in the menu).
- Under Project, check that it's the project you want the AI to work on.
- Untick Let AI manage everything in this project. That leaves only Read project and status, which is always on.
- Open Customize permissions and tick only what the job needs. The count next to it shows how many of the 12 you've picked.
- Under Token lifetime, pick 1, 4, 24, 72 or 168 hours.
- Press Create token, then Copy the AI message and paste it into your AI's chat.
For a token with fewer permissions, the ready-made message tells the AI which permissions it has and which it lacks. If the job needs one it doesn't have, the AI comes back and tells you instead of trying and failing. Only the account owner can issue a token with Read all app settings.
Permission sets that cover most jobs
- Have the AI read the logs and find the cause, without fixing anything yet: Read project and status alone, for 1 or 4 hours.
- Change some pages and publish again: add Upload files and check progress, and Publish the app.
- An app with a database: on top of that, add Create DBs, run SQL and retrieve credentials, Set secrets such as DB_PASSWORD, and Run app commands if it needs migrations.
- Domains or email: add Manage domains and verify DNS or Manage mailboxes, only for the job that needs it.
The platform already writes your database connection settings, such as DATABASE_URL, for your app, which reads them while it runs without the AI ever knowing the values. Most jobs don't need Read all app settings at all.
Your AI sends a code: three things to check before you approve
If your AI opens the project link without a token, the page tells it to ask for a pairing code, and the AI sends you a 10-character code with a link to the Connect AI page. Since 5 November 2026 you have 24 hours to approve a code, and the token it gives the AI lasts 3 days from your approval.
Press Enter AI code on the Overview, or use the box at the top of the Connect AI page, and type or paste the code. There's no project to choose, because the code already names it, and the review opens as soon as all 10 characters are in.

- Project requested: is it the project you asked the AI to work on?
- Permissions asked: do they stay within the job? You can't remove any on this screen, and AIs don't all ask for the same set; some include Read all app settings. If it's more than you want to give, don't press Approve. Ask the AI for a new code with the list of permissions you're willing to grant, or issue a token yourself as in the section above.
- You really asked this AI. If you didn't, do nothing: the code expires on its own.
Once you've approved, tell your AI "approved, go ahead". It picks up its token within seconds.
A message for your AI about secrets
Send this in the same chat where you pasted the link and token, and change the setting names to match your site:
The BrewInTH token you have carries only the permissions this job needs. If you need another one, stop and tell me which permission and why.
About the project's app settings:
- You may list the names, but never print a real value in the chat, a log or any file.
- Ask for a value only when the job truly needs it, and tell me first which one you'll ask for.
- Never write a password or key into the code; read it from an environment variable.
- I'll enter LINE_CHANNEL_TOKEN myself on the App settings tab; have the code read it by that name.
- Never run commands that print every value, such as env or cat .env.
Revoke a token when you're done, or if you think it leaked
- Open AI connections in the menu and choose the project under Project.
- Open Existing access / revoke. Tokens from the console's buttons are named My AI assistant; tokens from a pairing code carry the name the AI gave. Under each name is its expiry date.
- Press Revoke on that token.
The token stops working at once. Work it had queued and any version it was putting online are cancelled, but the live site isn't touched. Schedules the AI set up keep running, because they belong to the project, not the token; delete them on the Schedules tab if you don't want them. Hide token only hides the box; it doesn't revoke anything.
If you think a key leaked as well, replace it with the provider that issued it, update the value on the App settings tab and publish again (how to change a secret).
FAQ
My AI says it can see the names but not the values. Is something wrong?
No. A token shows only names by default. An AI with Read all app settings can ask for values one at a time, and your app still reads all of them as usual while it runs.
Can my AI use a key without seeing it?
Yes. Have the AI write code that reads the value by name, such as LINE_CHANNEL_TOKEN, then enter the value yourself on the App settings tab and publish again. The AI needs neither of the two app-settings permissions.
I pressed the automatic button a few times. Do the old tokens still work?
Yes, for their full 24 hours. The button creates a new token every time without revoking the old ones, so revoke the ones you don't use under Existing access / revoke.
Can one token work on several projects?
No. A token works only on the project it was issued for. If your AI works on two projects, give it one token for each; you can then revoke them separately.
If I revoke a token, does the site my AI built go away?
No. Everything the AI finished stays as it is; the token just can't do anything more. To let the AI carry on, issue a new token and send the short message in the same chat (carry on when a token expires).
Summary
On BrewInTH an AI sees and does only what its token allows. App settings are hidden from AIs by default so a value doesn't slip into a chat by accident, but running commands, database access and publishing can still reach secrets. Give tokens to AIs you trust, match permissions and lifetime to the job, check the permissions before you approve a code, and revoke a token when you're done. The in-console guide shows the real screens for each step, and the 10-point security checklist for AI-built sites covers the rest of your site.
