Trust · Credentials

Keys stay out of the model’s reach

The model can ask a tool to act. It never sees the token the tool uses to do it.

Where a credential goes

1

You connect

You authorize an account through its own sign-in screen. Brigham.ai never asks for a password.

2

Stored encrypted

The resulting token is encrypted at rest. The database holds ciphertext.

3

Referenced by id

Tools and runs hold only an id that points to the credential.

4

Resolved in code

When a tool runs, trusted code resolves the id and makes the call. The value is never placed in the prompt.

5

Redacted on the way out

Tool results and audit rows are redacted before they are written.

Handling rules

At rest: Encrypted
In transit: TLS
Model context: Never. Not in prompts, not in tool results
Logs and audit: Never. Redacted before writing
Revocation: Each connection is revocable on its own and stops working immediately
Starting access: Read-only. Write scopes are a separate, explicit grant
AI provider keys: Read from the erp.io shell at run time, not stored in Brigham.ai

Private by default, revocable on your terms

Each person connects their own accounts. A personal connection belongs to that person. Shared connections are a separate, deliberate choice. You can see every connection, its scopes and its health, and remove any of them in one click.