GovAI
Help center
Support that behaves like the product
Search across help and documentation, follow guided triage paths, and copy the same commands we use in CI. Product reference lives in Documentation.
Status checks
Quick mental model: health means the process is up; readiness means dependencies (for example Postgres) are satisfied.
HTTP surface — use /health vs /ready deliberately.
Local checks — curl probes and repo diagnostics from your machine.
Artefact gate — digest + export cross-check, not smoke-only.
Contact — pilots, procurement, or a suspected incident.
Browse by topic
Each card opens the most relevant help or documentation page.
What are you trying to do?
Pick a situation — we route you to the shortest path.
Interactive troubleshooting
Interactive troubleshooting
Symptom → verify command → expected signal → fix. Links jump to deeper docs.
Likely cause: Missing required evidence types and/or unmet approval / promotion prerequisites.
Verify
Expected: Structured JSON with non-empty missing_evidence or blocked_reasons
Next step: Replay the auditability scenario vocabulary for your policy pack, then append the missing events (see benchmarks/auditability-failures).
Animated explainers
Lightweight flows auto-advance — pause anytime. These mirror the same stages we reference in docs.
CI failure triage
Treat compliance-summary JSON as primary; logs only narrow the search space.
Evidence lifecycle
How an evidence event moves from emission to an auditable ledger record.
Compliance verdict flow
Prerequisites that must be satisfied before VALID is returned.
Export pipeline
From run_id to verifiable audit artefacts for buyers and regulators.
Deployment readiness
Binary up is not enough — /ready must reflect database and migrations.
Try consoles
Check audit service health
Differentiate liveness (process up) from readiness (dependencies such as Postgres).
Next stepLocal checks guide
Repository diagnostics
Run the same maintenance targets CI expects before a governance change lands.
Next stepContributing
Still stuck?
Send run_id (redacted), verdict JSON, and the failing command. For incidents include approximate time window (UTC).