Docs / Operations Operate
Monitoring
Health checks, growth metrics and what to alert on.
| Signal | How |
|---|---|
| Console liveness | GET /api/health (no sign-in; the image's HEALTHCHECK and the Helm chart's probes use it) |
| One project's health | ekos doctor --json, or GET /api/workspaces/<id>/health in the console |
| Console process | ekos console status, ekos console logs |
| Runs | each pipeline run's JSON-lines run log: steps, counts, warnings, model calls, outcome (Logging); the console's run page shows it as a steps table |
| Model spend | ekos llm usage (calls that reached a model, tokens and cost by task and model); the llm line of each run log |
| Ledger size | ekos status, ekos ledger status |
| Growth over time | ekos ledger timeline (JSON, cumulative counts by day, week or month) |
| Freshness | ekos freshness and ekos status: files changed since the last commit; FRESH001 at compile/commit |
| Silent gaps | ekos coverage --strict exits non-zero when an input kind produced zero objects |
| Reference integrity | ekos check refs --fail-on error (SARIF for code-scanning tools); the console's Integrity tab |
| Search index health | ekos ledger repair reports and heals |
| Pipeline warnings | .ekos/diagnostics/*.log; codes in Error Codes |
| Query activity | .ekos/query-log.jsonl |
| Agent access | the tenant's audit log in the console (Admin): who called which MCP tool, allowed or denied |
| MCP liveness | an ekos_status call through the server |
Suggested alerts: /api/health failing; doctor non-OK; a run ending in failure or not having run within its interval; object count drops (an unintended exclusion) or jumps (contamination); a spike in diagnostic warnings; new integrity errors; a scheduled backup failing.