Feature

Inspector.

Real-time observability into every AI interaction. Latency, token usage, policy hits, and cost — visualized in live dashboards with full drill-down. Search by person, team, model, or policy outcome.

What you see

  • Live feed — every request as it lands. User, model, latency, cost, policy outcome, receipt link.
  • Per-user view — what one person did this week. Time series of spend, token volume, blocked / rerouted / rewritten counts.
  • Per-team rollup — aggregate by department, project, cost center.
  • Model mix — which models are getting the most traffic; cost-per-task-completion estimates for routing decisions.
  • Policy heat map — which policies fire most, which violate the most, where the friction is.

What you can answer

  • "Who used Claude Opus this month and on what?"
  • "Which team is responsible for that $8K spike?"
  • "Did engineering's PII rate go up after the new model was added?"
  • "How often does our HIPAA policy block, and is that increasing?"
  • "What's the cost-per-completion for support tier 2 across providers?"

Inspector is the answer-engine for the question "what are we actually doing with AI?".

Drill-down

Click any event and see: full request metadata, policy decision tree (which rule fired, why), response metadata, signed receipt, cost accounting, and the upstream provider's per-model invoice line. The drill-down ends at the receipt; that's the auditable artifact.

Export

  • SIEM forwarding — native syslog, Splunk HEC, Datadog, Elastic, Wazuh
  • Notifications — Slack, PagerDuty, ServiceNow, email, webhook
  • Streaming API — SSE event stream for custom dashboards
  • CSV / JSON — on-demand for any time range

Adjacent reading

  • Evidence bundles — what compiles into auditor-ready archives
  • Fleet — the multi-tenant, multi-region picture
  • Router — turning insights into automatic routing decisions