For the complete documentation index, see llms.txt. Markdown versions of all docs pages are available by appending .md to any docs URL.
Observability
Monitor agentgateway with metrics, distributed traces, and access logs.
Agentgateway exposes three observability signals out of the box: metrics, distributed traces, and access logs. You can use these signals individually or together with tools like Prometheus and Grafana to monitor traffic health, LLM performance, and MCP tool usage.
MetricsPrometheus-compatible endpoint for request, LLM, MCP, and connection metrics
TracesOTLP distributed tracing with OpenTelemetry semantic conventions for LLM and MCP traffic
Access logsPer-request structured logs with CEL-based filtering, enrichment, and OTLP export
Grafana dashboardImport the pre-built dashboard to visualize request, LLM, MCP, and connection metrics
Observability by traffic type
Different traffic types expose additional signal-specific fields.
- LLM traffic: token counts, model names, provider, cost, time to first token, and time per output token are available in all three signals. See Observe LLM traffic.
- MCP traffic: session ID, method, server, and resource are available in traces and metrics. See Observe MCP traffic.
External LLM observability platforms
For specialized LLM observability platforms with prompt logging, cost tracking, and evaluation capabilities, such as Langfuse, Langsmith, or Phoenix, see the LLM Observability integrations.
Was this page helpful?
