Tracing

Tracing the execution of LLM applications

A simplified view of a trace containing spans

Tracing is a powerful tool for understanding how your LLM application works. It helps you track down issues like application latency, runtime exceptions, incorrect prompts, poor retrieval, and more.

Arize is built on top of open source packages OpenTelemetry and OpenInference. We have native support for the OpenTelemetry protocol, which is vendor-agnostic, open source, and highly performant. It includes batch processing that can handle billions of traces and spans. OpenInference standardizes traces and spans data across models, frameworks, tool calls, prompts, retrievers, and more.

It takes only a few lines of code to setup tracing using Arize with our auto-instrumentation, and it's flexible enough to add your own metadata and customize your spans.

A full example trace within Arize

To get started with code, check out the Quickstart guide for LLM tracing and evaluation.

Learn more about tracing concepts by reading our articles on What are Traces? and How does Tracing Work?

Deciding between automatic vs. manual instrumentation? Read how to setup tracing.