Ragas
A Python library, Apache 2.0 licensed, for evaluating model-based applications, RAG systems first. It computes judge-scored or classic measures: faithfulness of the answer to the context, answer relevance, context precision and recall (recall needs a reference answer per case), factual correctness, noise sensitivity, plus measures for agents and SQL. It can also generate a test set from your documents. The repository moved organization on GitHub (explodinggradients became vibrantlabsai, Vibrant Labs). Latest release: 0.4.3, on January 13, 2026; no contribution merged since February 24, 2026, while 271 pull requests remain open as of November 12.
Best for
- Telling whether a bad RAG answer comes from retrieval or from generation
- Generating a first set of test questions from your own documents