Skip to main content
How It WorksSandboxPricing
← All features

REASONING RECORDS

Intent Chain

Capture why an agent made each decision — AES-256 encrypted, hash-chained, forensic-grade proof of reasoning.

How it works

The Intent Chain captures the agent’s reasoning before each action. The reasoning text is encrypted client-side using AES-256-GCM before transmission — raw reasoning never leaves the agent process unencrypted. The governance system stores only the encrypted payload and its hash. The intent_hash is included in the event_hash computation, making the reasoning tamper-evident. Decryption is restricted to authorized users and every decryption request is logged.

Technical details

Encryption: AES-256-GCM, client-side before transmission. IV: 12 bytes random per event. Auth tag: 16 bytes. Storage format: iv:tag:ciphertext (hex:hex:base64). intent_hash = SHA-256(packed ciphertext). Intent coverage = (events with intent / total events) * 100.

Compliance mapping

ISO 42001 6.4 (AI Risk Treatment)EU AI Act Article 13 (Transparency)

Use cases

  • Proving an agent’s decision was based on correct inputs
  • Regulatory investigations requiring decision audit trails
  • Insurance claims requiring proof of agent reasoning

Related features

Hash ChainAgent CertificateHuman Accountability
Start governing your agents →Read the standard →