TruLens

TruLens is an open-source library for tracing and evaluating AI agents and RAG apps. It records each step with OpenTelemetry and scores it with LLM judges that explain their reasoning, helping developers compare versions and decide which one to ship.

Every span captures inputs, outputs, latency, tokens and cost, so a bad answer can be traced to the step that caused it. Built-in judges cover groundedness, context relevance and answer relevance, known together as the RAG triad, and recent versions can evaluate full multi-turn conversations.

You can tune a judge to your domain by adding your own rubric, few-shot examples and score range in Python. The project is maintained by Snowflake and published on GitHub with public documentation.