Is there a way to audit exactly *why* a model made a specific tool call?

Standard observability tools like LangSmith or Helicone show you the latency, tokens, and the prompt that was sent. But they don't capture the underlying business state or policy decision that authorized the action.

If an agent deletes a user account, you need to prove the user had the correct role at that exact timestamp.

Exogram's Trust Ledgers log the evaluation_id, the payload_hash, and a context_snapshot of the graph state. This immutable ledger proves not just what the AI requested, but why the deterministic policy engine allowed it, providing absolute non-repudiation for enterprise deployments.