Prometheus¶

Connect HolmesGPT to Prometheus for metrics analysis and query generation. Enables detection of memory leaks, CPU throttling, queue backlogs, and performance issues.

Prerequisites¶

Prometheus server running and accessible
Network connectivity from HolmesGPT to Prometheus endpoint

Configuration¶

Holmes CLIRobusta Helm Chart

Add the following to ~/.holmes/config.yaml, creating the file if it doesn't exist:

toolsets:
    prometheus/metrics:
        enabled: true
        config:
            prometheus_url: http://prometheus-server:9090

            # optional
            #headers:
            #    Authorization: "Basic <base_64_encoded_string>"

holmes:
    toolsets:
        prometheus/metrics:
            enabled: true
            config:
                prometheus_url: http://prometheus-server:9090

                # optional
                #headers:
                #    Authorization: "Basic <base_64_encoded_string>"

Update your Helm values and run a Helm upgrade:

helm upgrade robusta robusta/robusta --values=generated_values.yaml --set clusterName=<YOUR_CLUSTER_NAME>

💡 Alternative: Set PROMETHEUS_URL environment variable instead of the config file.

Validation¶

Test your connection:

holmes ask "show me CPU usage for the last hour"

Troubleshooting¶

Finding Your Prometheus URL¶

# Find Prometheus services
kubectl get svc -A | grep prometheus

# Port forward for testing
kubectl port-forward svc/prometheus-server 9090:80
# Then use: http://localhost:9090

Common Issues¶

Connection refused: Check if Prometheus URL is accessible from HolmesGPT
Authentication errors: Verify headers configuration for secured Prometheus
No metrics returned: Ensure Prometheus is scraping your targets

Capabilities¶

Tool Name	Description
list_available_metrics	List all available Prometheus metrics
execute_prometheus_instant_query	Execute an instant PromQL query
execute_prometheus_range_query	Execute a range PromQL query for time series data
get_current_time	Get current timestamp for time-based queries