AUDIT-READY LOGGING

Trace every AI decision

Decision Trace captures the full context of every AI decision—inputs, sources, citations, tool actions, and outputs—making automation reproducible, auditable, and improvable.

See Decision Trace in action

Reproducible

Replay any decision with the exact same inputs to understand what happened.

Auditable

Export complete decision history for compliance, legal, and regulatory needs.

Improvable

Use traces to identify failure patterns and improve prompts and workflows.

Measurable

Track cost, latency, and quality metrics across all interactions.

Decision Trace schema

Every interaction captures these minimum fields. Additional fields are configurable per deployment.

decision_trace.json

{
  "interaction_id": "uuid",
  "timestamp": "2026-01-05T14:30:00Z",
  "channel": "voice",
  "customer_id_hash": "sha256...",

  "prompt_version": "v2.4.1",
  "policy_version": "v1.2.0",
  "model": "gpt-4o",

  "retrieval_sources": ["policy_v1.2.pdf", "faq.md"],
  "citations": [{"source": "policy_v1.2.pdf", "page": 3}],

  "tool_calls": ["check_availability", "create_booking"],
  "tool_outputs_summary_hash": "sha256...",

  "decision": "book_appointment",
  "confidence_score": 0.94,
  "risk_level": "low",

  "approval_required_flag": false,
  "final_outcome": "success",

  "duration_ms": 1240,
  "model_cost_usd": 0.0034,
  "infra_cost_usd": 0.0001
}