Telemetry

Alexandria exports traces, metrics, and logs via OTLP HTTP (port 4318) when an endpoint is configured. Zero overhead when disabled.

Enable

alexandria config set otel.endpoint "http://localhost:4318"

Optional: change service name or sampling rate.

alexandria config set otel.service_name "my-alexandria"
alexandria config set otel.sample_rate "0.5"   # sample 50% of traces

Check status:

alexandria telemetry status

Run the Collector

Download otelcol-contrib and run with the bundled template:

otelcol --config config/otelcol.yaml

The default config exports to stdout (debug mode). Uncomment the relevant exporter block in config/otelcol.yaml (in the source tree) for your backend:

• Jaeger — otlp/jaeger
• Grafana Tempo — otlp/tempo
• Prometheus — prometheusremotewrite
• Datadog — datadog

Signals

Signal	What's captured
Traces	One span per query, LLM completion, and tool invocation
Metrics	query.count, query.duration_ms, llm.duration_ms, tool.duration_ms, tool.cache_hits, tool.cache_misses
Logs	All info/warn/error events correlated with trace context

All metrics carry a backend.name or tool.name attribute for filtering.