Cadrena is building the authorization and evidence layer for AI agents: authorize before execution, preserve evidence after.

_{Building in public. We are validating the first production authorization slice.}

AI agents can read customer data, change production systems, move money, and call tools on a user's behalf. Existing identity systems grant broad access; they do not reliably decide whether one parameterized agent action should run right now.

Cadrena is an open-source control layer for that decision:

agent action → policy decision → approval when required → execution → evidence

• Runtime authorization before an MCP tool call reaches the upstream system.
• Parameter-aware policy for the actual action, resource, and execution context.
• Bound approvals that cannot be replayed against a changed invocation.
• Durable evidence for every decision and execution outcome.

• Enforcement before observation. Stop unsafe actions before they happen.
• Deterministic boundaries. Models may supply context; they do not grant authority.
• Least privilege by default. Unverified or ambiguous context never widens access.
• Approval integrity. Approval is bound to the exact action and policy state.
• Evidence by construction. Every decision produces an explainable record.

We are validating Cadrena component by component. Repositories will appear here as they are renamed, documented, and ready for external use.

If you are working on production agents, MCP infrastructure, authorization, or security tooling, follow Cadrena and check back as the system takes shape.