Phoenix Tracing

This is your reference guide for adding Phoenix observability to LLM apps. It covers both Python and TypeScript implementations, walking you through everything from basic setup with auto-instrumentation to creating custom spans for the nine OpenInference span types (LLM, chain, retriever, tool, agent, embedding, reranker, guardrail, evaluator). The production guidance on batch processing and PII masking is genuinely useful if you're shipping this to real users. The documentation is organized well with clear prefixes, so you can jump straight to span attribute schemas or session tracking patterns without reading everything. Assumes you have a Phoenix server running and are already somewhat familiar with observability concepts.