Five steps. Every agent action. Every time.
ProvenanceCode intercepts every AI agent decision before it executes — evaluating it against policy, generating a signed provenance record, and routing high-risk actions to a human approver. Nothing runs without passing through.
Your AI agent reasons about a goal and produces a structured decision — what it intends to do, and why. It does not execute directly.
Agent OS receives the decision before it reaches any executor. The action is paused. Nothing runs yet.
The decision is evaluated against your policy rules — ISO 42001-aligned by default. Compliant actions proceed. Non-compliant actions are blocked outright or gated for review.
High-risk actions — production deployments, data deletions, financial operations — are paused and routed to a designated approver via SMS or email. Execution resumes only on explicit approval.
A cryptographically signed record is created for every execution event — before and after. This is your audit trail: immutable, attributable, and replayable.
Zero-trust execution
Executors trust the provenance record and policy — not the agent. The agent reasons; the executor acts. The two are never conflated.
Deterministic policy
Rules are YAML. Evaluation is deterministic rule-matching — not probabilistic inference. You can read, audit, and version-control every policy decision.
Complete audit trail
Every action — approved, blocked, or gated — produces a signed, timestamped provenance record. Your board, regulator, and insurer can see exactly what ran and why.
Deploy in minutes. Govern immediately.
Available on AWS Marketplace. Two inputs: your LLM credential + your ProvenanceCode API key. ISO 42001 policy pre-loaded.