This doc focuses on the "ops hardening" layer: logs, metrics, and safe execution defaults.
The backend emits structured JSON logs to stdout/stderr.
Correlation:
request_id: provided byX-Request-Idheader (or generated automatically)run_id: set when a run is created and used for tool/policy/approval logs
Env:
RUNVANE_LOG_LEVEL(defaultINFO)
Backend exposes:
GET /metrics
Currently returns a JSON snapshot with:
- uptime
- in-memory counters for tool invocations/success/failure/circuit-open
Tool invocation is wrapped with:
- retries per tool (
ToolSpec.retries) - a per-tool circuit breaker after consecutive failures
Env:
RUNVANE_CB_FAILURE_THRESHOLD(default3)RUNVANE_CB_RESET_SEC(default60)
Before tool results are stored in traces/UI/logs, Runvane performs best-effort redaction for:
- bearer tokens
sk-...-style secrets- private key blocks
Pending tool inputs (used after approval) are not redacted so execution still works.