TruLens

TruLens is an open-source Python evaluation framework originally created by TruEra and now maintained by Snowflake following its acquisition in May 2024. TruEra was founded by ML interpretability researchers, and that lineage is visible in TruLens's core abstraction: feedback functions that attach to any input, output, or intermediate trace step and return explainable float scores rather than opaque pass/fail results. For teams running on Snowflake Cortex, TruLens is the native evaluation layer, with traces logging directly to Snowflake and feedback functions doubling as production guardrails.

• Feedback functions and Metric class API: Programmatic evaluators that score inputs, outputs, and intermediate steps on a 0.0–1.0 scale, with chain-of-thought reasoning support for interpretable results

• RAG Triad built-in metrics: Context Relevance, Groundedness, and Answer Relevance, a three-metric evaluation framework for RAG pipelines that has been widely adopted beyond TruLens itself

• Snowflake Cortex native integration: Traces log directly to Snowflake; feedback functions run against Cortex Search and Cortex LLM Functions without additional configuration

• Stack-agnostic instrumentation: Works with LangChain, LlamaIndex, LangGraph, and raw Python across OpenAI, Anthropic, HuggingFace, and Snowflake Cortex as feedback LLM providers

• OpenTelemetry-native tracing: OTel span support makes TruLens trace data portable to Datadog, New Relic, and other observability stacks

• Built-in local dashboard: Visualizes feedback scores and traces across evaluation runs via session.start_dashboard(), with no external service required