3AM Doctor
Is 3AM itself healthy? Every check (core, licence, intake, models and their placement, connections, approvals, accuracy, the audit log, disk) with what to do, in the console, the CLI and the API.
3AM Doctor (/doctor, g h) gives the running install's health, check by check, each with what it found and what
to do. The overall status is the worst of its checks: ok, warn or fail. Start here when something seems off,
before you call anyone.
./install.sh --doctor # single server
podman exec 3am-core /opt/3am/3am-core doctor --json # the same, as JSON
kubectl -n bank-ops exec statefulset/threeam -- /opt/3am/3am-core doctor
curl -s -H "Authorization: Bearer $T" "$URL/v1/doctor?fresh=1"| Check | Fails or warns when |
|---|---|
| Core and the licence | The licence is missing, lapsed, unconfirmed (strict), or in grace |
| Alert intake | Alert sources aren't polled, or a source keeps failing |
| What changed | Snapshots can't be read |
| Investigations | Investigations fail or stall |
| Every connection | Its live test fails |
| Model placement | A host without a GPU whose models are missing or local (fail) |
| System 1 and the investigator | Either doesn't answer |
| Approvals | Requests waiting; an escalation step whose channel isn't connected; the lab auto-approver is on (fail) |
| Policy | The fix mode, and the quorum for raising autonomy |
| Accuracy | The share of judged diagnoses that were right |
| The audit log | Its hash chain or a seal doesn't verify |
| Disk | The data volume is filling up |
Export: a PDF report (/doctor/report), the checks as CSV, or copied as text for a ticket.
Health and monitoring has how to watch 3AM from your own monitoring.