Skip to content

Monitoring

Every run appends a UsageRecord to a per-tenant usage log (.riptide_watergraph/usage.jsonl) with tenant, mode, tokens, cost, latency, success, tool-call counts, and a timestamp.

In the Studio

riptide serve → Monitoring aggregates the log into:

  • KPI cards — runs, success rate, avg latency, tokens, cost, tool-call validity, blocked
  • a runs / cost-over-time chart (inline SVG, no libraries)
  • a recent-runs table

served by GET /api/monitoring (totals, by_mode, by_tenant, recent, daily).

Costs

riptide costs

prints the per-tenant dashboard (runs, tokens, cost, blocked). Per-tenant budget ceilings reject over-budget runs with HTTP 402.

Deeper tracing

The optional [observability] extra adds per-LLM-call spans via OpenTelemetry + Langfuse. Tracing is opt-in and off by default (RIPTIDE_WATERGRAPH_DISABLE_TRACING=1 in the Docker image).

pip install "riptide-watergraph[observability]"