Access is evaluated for each request
An agent receives only the categories, items, and fields allowed by its active grant and your current policy. Paused, expired, revoked, and locked grants are denied.
Security model
Your data stays under your control.
Limits is designed to give AI agents narrow, revocable access to the information you choose to share.
Sensitive values use explicit approval
Sensitive and restricted information can require your approval. Fill-only values are submitted into a compatible form without being returned as readable text to the agent.
User-held releases
The connected client must register an agent session key before approved fields can be released. Limits sends only ciphertext encrypted for that key and never decrypts user-held values for an agent. OTP requests return a current code; the seed is never released.
Review every access decision
The activity log is designed to show which agent requested access, which item it requested, why, and what decision was made.
Current preview status
This preview uses account sign-in, encrypted vault storage, and live access control. Test-mode cards do not issue a bank card or send a payment.