Capture evidence
Import an incident or ingest an OpenTelemetry trace. Pin the logs and spans that establish what happened.
Agent evaluation platform
Turn logs and traces into test cases. See whether your investigation agent finds the right cause, cites the evidence, and knows when to ask for more.
A workflow for engineers who build and evaluate investigation agents.
Import an incident or ingest an OpenTelemetry trace. Pin the logs and spans that establish what happened.
Replay a versioned dataset against two candidates. Compare their diagnoses, citations, latency, and cost on matched cases.
Use the scoring SDK and GitHub Action to flag regressions, including missed approvals and unsupported security conclusions.
A/B comparison
In this recorded example, both agents see the same three incidents. Only Guarded v3 asks for a server trace when the logs cannot explain the timeout.
| Incident | Triage v1 | Guarded v3 |
|---|---|---|
| Forecast provider timeout | Pass | Pass |
| Forecast response parsing | Pass | Pass |
| Unknown client timeout | Fail | Pass |