Trace LLM calls with OpenTelemetry to debug your AI agent

Short summary

A practical guide to instrumenting LLM calls with OpenTelemetry so you can trace prompts, responses, timing, and cost. The author walks through wrapping existing API calls with minimal code changes using OpenInference instrumentation libraries. Several observability platforms are surveyed including LangSmith, Langfuse, and the open-source Arize Phoenix.

• •Use OpenTelemetry to record every LLM call's prompt, response, timing, and cost
• •Instrumentation requires just two lines wrapping existing model API calls
• •Tools surveyed: LangSmith, Langfuse, Helicone, Arize Phoenix (open source), Enprompta

Generated with AI, which can make mistakes.