White paperGoverning AI agents in production: policies, human approvals and an emergency stop.Read the white paper
NavinAI Platform
FRFrançais
Documentation
Request a demo

Traceability

Prove what every agent did, and why

Every call carries a W3C traceparent and produces OpenTelemetry spans, a Usage Event and a hash-chained audit event. Evidence is kept in WORM storage.
Request a demoContact us

At a glance

  • Hash-chained audit, tenant by tenant
  • Evidence in WORM storage (Object Lock)
  • Replay of every run

The challenge

A decision made by an agent must be explainable

Regulators, auditors and business owners ask the same questions: who asked, which model answered, with which data, under which policy, at what cost. Scattered logs cannot answer them.

  • Logs that can be altered, and therefore disputed
  • Traces broken between services
  • No link between an answer, the data used and the policy applied
  • A run that cannot be reproduced

Chained audit

Every event seals the previous one

Each audit event contains the hash of the event before it. Changing or deleting an event breaks the chain, and the break is detected.

  1. Event n-1

    Hash h(n-1)

  2. Event n

    Content + h(n-1) = h(n)

  3. Event n+1

    Content + h(n) = h(n+1)

  4. WORM storage

    Object Lock: no change, no deletion

The identity fields of the audit record who acted, on behalf of whom and under which mandate.

From trace to evidence

Everything needed to explain a run

  • End-to-end traces

    The W3C traceparent follows the request through every gateway and runtime; OpenTelemetry and the OTel Collector carry the spans.

  • AI tracing

    Opik records prompts, answers, tools called and evaluations, in line with DLP rules.

  • Usage Events

    Every call produces a usage event: model, tokens, duration, cost estimated then measured.

  • WORM audit evidence

    Audit events are kept in object storage with Object Lock and correlated across services.

  • Execution replay

    An agent run can be replayed from its checkpoints to understand, fix or demonstrate.

  • Data lineage

    OpenLineage links every dataset to its sources and to its use by pipelines and agents.

  • SIEM export

    Audit and security events are exported to your SIEM; evidence follows a chain of custody.

  • Logs and analytics

    OpenSearch, ClickHouse and Prometheus keep logs, analytics and metrics for operations.

The proof

Traceability your auditors can use

Evidence is produced by the platform itself, as things happen, not reconstructed afterwards.

  • Integrity checks of the audit chain
  • Correlation of audit, traces and usage
  • Retention periods per data type
  • Access to evidence is itself audited

Go further

  • ArticleTraceability

    A tamper-evident audit trail for every AI call

    Every call that crosses a Navin Enterprise AI gateway carries a W3C traceparent and produces OpenTelemetry spans, a Usage Event and an audit event chained by hash. Here is how the chain is built, sealed, archived in WORM storage and replayed.

    3 min read

The other capabilities

Get ready for your AI audits today

We show you how an agent run is traced, from the request to the evidence.

Request a demoContact us