Verb-level risk controls
Every integration action and platform tool is classified LOW, MEDIUM, or HIGH. Risk is enforced at the action level — not per conversation.
Approval gates & policies
Define what's allowed. The system enforces it on every execution — tenant-wide or per Cerebral. Match on verb name, risk level, provider, or conditions like amount or recipient.
Shadow Mode & dry run
Run every operation without executing it. Approve behavior before it ever touches production. Both are built into the same execution path as live mode — not separate systems.
Immutable audit trail
Every action, approval decision, and escalation logged with tool name, inputs, outputs, duration, and execution ID. Always on. Not configurable off. Built for compliance teams.
Idempotent execution
Every write operation is assigned a unique execution signature before it leaves the platform. If a retry fires — from a timeout, a restart, or a concurrent duplicate request — the fence catches it at the constraint level and returns the cached result. The third-party API is never called twice. Enforced at the infrastructure layer across all 2,800+ providers, regardless of whether they natively support idempotency keys.