For the complete documentation index, see llms.txt. Markdown versions of all docs pages are available by appending .md to any docs URL.

View, customize, and reference data plane metrics emitted by the agentgateway proxy.

The agentgateway proxy exposes a Prometheus-compatible metrics endpoint on port 15020. Metrics are collected automatically for every request that passes through the gateway and follow the OpenMetrics format. All agentgateway metrics use the agentgateway_ prefix.

To set up automatic scraping of these metrics with Prometheus, see Enable metrics scraping.

View data plane metrics

1. Port-forward the agentgateway proxy.

   kubectl port-forward -n agentgateway-system deploy/agentgateway-proxy 15020:15020

2. Query the metrics endpoint.

   curl http://localhost:15020/metrics

3. Enable metrics scraping for data plane metrics with the OTel stack so that you can export and visualize metrics in monitoring tools, such as Prometheus and Grafana. For more information, see Scrape metrics for querying and visualization.

Add custom metric labels

You can enrich all metrics with custom labels computed from CEL expressions by using an AgentgatewayPolicy. Labels are added to every metric that carries route identifier labels.

The following example adds two labels to all metrics:

• team: A dynamic label computed from a JWT claim, so each metric reflects the team that owns the request.
• org: A dynamic label read from the x-org-id request header.

kubectl apply -f- <<EOF
apiVersion: agentgateway.dev/v1alpha1
kind: AgentgatewayPolicy
metadata:
  name: metrics-labels
  namespace: agentgateway-system
spec:
  targetRefs:
  - group: gateway.networking.k8s.io
    kind: Gateway
    name: agentgateway-proxy
  frontend:
    metrics:
      attributes:
        add:
        - name: team
          expression: jwt.team
        - name: org
          expression: 'request.headers["x-org-id"]'
EOF

Data plane metrics reference

XDS

These metrics track communication with the agentgateway control plane over XDS. They are only present when agentgateway runs in Kubernetes mode and do not appear on a standalone process.

HTTP

TCP

MCP

LLM

These metrics follow the OpenTelemetry semantic conventions for generative AI.

Cgroup memory

These metrics track Linux cgroup v2 memory statistics for the agentgateway process. They are available whenever agentgateway runs in a Linux environment, including standalone deployments, and are not specific to Kubernetes mode.

Process memory

These metrics track process-level memory for the agentgateway process, sourced from /proc/self/smaps. They are available whenever agentgateway runs in a Linux environment, including standalone deployments, and are not specific to Kubernetes mode.

Misc

The following metrics reflect the health of the agentgateway process and appear on every startup regardless of traffic.

Example PromQL queries

Use these queries in Prometheus or as the basis for Grafana panels and alerts. To enter a raw PromQL query in Grafana, switch the query editor from Builder to Code mode.