instrumenting-with-mlflow-tracing

Trace LLM calls and tool execution in MLflow by instrumenting your code paths during workflow setup. This records prompt and response details, token usage, and external dependency calls, allowing you to verify traces appear by running your agent and checking with mlflow.search_traces().

Trace end-to-end agent workflow execution in MLflow to capture root operations for pipelines and agent steps. This provides visibility into where time is spent across your AI workflow, helping you quickly diagnose latency, errors, and quality issues.

You need to load MLflow tracing during workflow setup to ensure it captures end-to-end workflow execution. Tracing must be initialized as part of the pipeline configuration to successfully record root operations, LLM calls, and tool interactions.

Trace retrieval and tool calls in MLflow to debug agent workflows by recording retrieval activity and external dependency calls. This resolves blind spots in AI pipelines by capturing prompt and response details alongside token usage for debugging.

Instrument TypeScript code paths with MLflow tracing to provide observability for AI agent workflows. It captures LLM calls, tool execution, and agent routing decisions across both Python and TypeScript implementations, supporting framework instrumentation.