Open-source framework for evaluating and tracking LLM application quality. By 2026, TruLens (from Snowflake/TruEra) is widely used for groundedness, relevance,
TruLens brings feedback-function evaluation to LLM apps: programmatic scoring of groundedness, relevance, and custom criteria across traces, the open-source path (now in the Snowflake orbit) for teams instrumenting RAG quality with the 'RAG triad' methodology.
Pricing: Open-source free; managed capabilities surface via Snowflake Cortex ecosystem.
Three feedback scores: context relevance (did retrieval fetch the right material), groundedness (is the answer supported by it), and answer relevance (does it address the question), which together localize RAG failures precisely.
TruLens leans runtime feedback scoring inside apps; promptfoo leans declarative pre-deploy testing in CI. They cover complementary points of the quality lifecycle.