Trust & security

Enforced in the database, not in prompts.

Every agent runs inside a bounded, human-governed lane. The controls below are the ones your security and compliance teams will ask about.

Isolation

Your data stays in your tenant

Every row carries a tenant ID, and row-level security enforces the boundary in Postgres itself — not in application code.

Isolation

Research agents can read, never act

Agents that read the open web hold read and search tools only. No agent both ingests untrusted content and acts on the world.

Integrity

Your outcome history is never rewritten

Database triggers reject every update and delete on outcome records; corrections are compensating rows. Your history can't be rewritten by anyone, including us.

Control

Every run is bounded

Every autonomous run carries hard per-run and per-account limits and a bounded loop. Crossing a limit ends the run cleanly and logs why.

Compliance

Every send passes a gate

Opt-outs and suppressions have dedicated ledgers and dispatch controls. Consent, reply, bounce, complaint and suppression gates are all checked before a message leaves Cernu.

Privacy

Export and erasure on request

Tenants can export their data and erase a contact or an account. Personal data in append-only ledgers is held by reference, so it can be severed on request without breaking the outcome record.

Assurance

Every schema change is on the record

The migration ledger is append-only and checksum-verified, and every hosted change is read back over verified TLS before it counts. Nothing reaches your data unrecorded.

Vendors

Subprocessors

The list is short: cloud infrastructure, model inference, data enrichment, and telephony. Each vendor key holds only the capability the job needs.

Bring your security team.

We'll walk them through the real schema, triggers, and policy gate — not a diagram of them.