Stop unsafe AI-agent actions before they execute.
Imladri sits between an agent and its privileged tools, enforces a versioned policy, and produces a verifiable chain-of-custody proof packet for every governed decision.
Chain-of-custody evidence
Watch the block happen before the action body.
Run a policy-cache decision or call the live runtime preflight. The browser receives redacted evidence and an inspectable proof digest.
Live policy cache block
Loads the live D1 policy row, then runs the same local allow/block semantics used by warm SDK adopters.
pendingpendingsha256:pendingpendingmsOne policy decision becomes one verifiable proof chain.
One agent, three boundaries
Denied before the function body runs.
A governed Postgres branch runs without mutating the source.
The compute run emits attestation, digest, and cleanup evidence.
Three boundaries on the record.
Terminal and API wrappers block prohibited attempts before body entry, then sync the same evidence shape into Profile.
Governed Postgres branches prove source isolation under the same policy and packet model.
Protected GPU runs add attestation, digest evidence, and zeroized cleanup to the same record.