Security - Brea

The security behind Brea.

Here is exactly how your data, your keys, and your account stay protected.


01 / Credentials

Your keys never meet the model.

Passwords, API keys, and tokens are stored encrypted and injected server-side at the moment a tool runs, outside the machine your agent runs on. The AI works with references, never raw secrets.

Trace / tool call live

→

model requests

gmail.create_draft { account: "ref_8f2a…" }// no secret in context

→

broker verifies

owner ✓ · scope ✓ · pinned to your account // server-side

→

key injected

authorization: •••••••••••••••• // never enters the chat

←

result returned

201 · draft ready for your review // key stays behind

Fig. 01: one real tool call. A reference is all the model ever holds.

So even if the model is asked to reveal a key, it has nothing to reveal.


02 / Data handling

What Brea does. What it never will.

Brea always:

Brea never:


03 / Infrastructure

A private machine, just for you.

Your agent does not live in a crowd. It runs on its own isolated instance: provisioned when you join, destroyed when you leave.


04 / Compliance

Independently assessed.

CASA Tier 2 • App Defense Alliance •

Brea has passed the App Defense Alliance's Cloud Application Security Assessment at Tier 2, the security bar Google requires before an app may work with restricted account data like Gmail.

Brea is also a Microsoft Verified Partner.

Responsible disclosure

Found something? Email security@brea.ai. A human reads every report.


05 / Questions

Asked, answered.

Can the AI see my passwords or API keys?

Is my data used to train AI models?

What is CASA Tier 2?

What happens when I disconnect an app?

What happens to my data if I cancel?

How do I report a security issue?

More questions? Email us


Meet Brea

By now your agent could already have started.

Your to-do list, handled. Connect your tools, send a text, and let Brea take it from there.