Trulens

Open-source trace and evaluation framework for LLM applications across any framework.

Overview

Trulens decouples observability from your application architecture, offering decorators and callbacks that automatically capture execution traces, latency, and token usage. It pairs tracing with built-in evaluation metrics, allowing developers to score agent behavior and RAG quality directly in their codebase.

Best for

• framework-agnostic tracing
• custom evaluation metrics
• local-first observability

Trade-offs

What XeroHack pre-wires

When you pick Trulens in the interview, the engine emits these files into your scaffold:

• @observe decorator for Python functions
• default RAG faithfulness/relevance checks