How CAUM Works

CAUM observes your AI agents structurally. It sees only tool names and timestamps — never prompts, outputs, or content.

What you see in the dashboard:

Regimes describe what the agent is doing:

• EXPLORER — Trying different approaches (healthy)
• STABLE — Consistent read-write-test cycle (healthy)
• GRIND — Repetitive but making some progress
• LOOP — Same action repeated without progress
• REASONING_LOOP — Circular reasoning, completely stuck

Certificates are cryptographic proofs. Each step is hashed into a Merkle chain. Click any session, then "View Certificate" to see the full attestation.

CAUM never decides, never blocks, never controls. It observes and attests.