Runs and Evidence

List the newest Runs:

curl --fail -H 'Authorization: Bearer example-token' \
  'http://127.0.0.1:8090/api/v1/runs?limit=20'

Fetch one Run with GET /api/v1/runs/{id}. Native API responses include run_id; /v1/chat/completions returns X-Akritas-Run-ID. Run list responses omit raw tool payloads, while authenticated Run detail includes host-recorded tool arguments and raw results. Protect Run detail as operationally sensitive data.

Use tool events as execution evidence. A Run marked succeeded means its host workflow completed; it does not guarantee the model answer is correct or an operational recommendation is safe.

Alert ingestion responses provide event_id, incident_id, and job_id. Poll GET /api/v1/investigation-jobs/{job_id} and follow its run_id, or query the incident directly through GET /api/v1/incidents/{incident_id}. A successful alert Run includes the validated structured investigation.

Use POST /api/v1/runs/{id}/chat for a bounded follow-up against a completed Run. The handler supplies the stored investigation and timeline to the normal read-only Chat loop and appends new tool and follow-up events to the same Run.