Logfire — Python Observability on OpenTelemetry

Practical Notes

• Quant: aim for 1 end-to-end request trace with spans for DB + HTTP within the first hour.
• Quant: track p95 latency and error rate before vs after adding instrumentation.

Agent backend tip: make tool calls observable

If you run an agent service, the fastest win is instrumenting tool calls:

• external HTTP requests,
• database queries,
• vector search,
• and model API calls.

Name spans consistently (e.g., tool.http, tool.sql, tool.vector) so you can answer: where did time and failures go?

Data hygiene

Avoid logging secrets. Configure scrubbing/redaction early and review what ends up in traces before enabling in production.

FAQ

Q: Is the whole platform open-source? A: No. This repo is the open-source SDK and docs; the hosted platform is separate.

Q: Can I export to another backend? A: Yes—OpenTelemetry compatibility means you can target other OTel backends if needed.

Q: What should I instrument first? A: Start with request traces + tool calls, then add DB/HTTP client instrumentation.