AI and models · Edition No. 14 · 8 Sep 2026

vibrantlabsai/ragas

Scores how good a retrieval-based AI answer is, so a change to Grasppy can be shown to have helped or hurt.

← AI and modelsRead the whole edition →

15.7k stars · Apache-2.0 · v0.4.3 (2026-01-13) — GitHub and PyPI agree

Scores how good a retrieval-based AI answer is, so a change to Grasppy can be shown to have helped or hurt.

Repo detailsthe review · specs · pros & cons · install

What it is

Ragas is an evaluation framework for AI applications that look things up before answering. It measures things like whether the answer is supported by the source text, and whether the right parts of the source were found.

Why it matters

Grasppy's entire pitch is the map. You have several ways to change how it is built. LightRAG (Ed. 6 #8), pgvector (Ed. 2 #11), BERTopic and turftopic all change the result. Without a score, choosing between them is guessing, and you will keep the one you happened to build last. Two earlier entries sit either side of this. Instructor (Ed. 5 #11) makes the output the right shape. Langfuse (Ed. 10 #7) watches what happened after release. This is the missing middle: a number before release that you can compare.

Stars15.7k
LicenceApache-2.0
Latestv0.4.3 (2026-01-13) — GitHub and PyPI agree
Good
  • It turns "does this feel better" into a number you can put in a table.
  • It works with what you already use, including LangChain and plain Python.
  • Apache-2.0 licence, so there is nothing to worry about in a paid product.
Watch for
  • Every run costs money, because a language model does the scoring. Keep the test set small, twenty to fifty examples, not thousands.
  • You have to write that test set yourself. That is the real work, and it is most of the day.
  • Scores from a language model wobble between runs. Compare one run against another. Never treat a single number as the truth.
Install
source venv/bin/activate
pip install ragas

Get the next edition in your inbox

A dozen repositories, opened and checked. The licence read, the last release dated, and the ones that did not make it named with the reason. It is the half most lists leave out.

No tracking pixels. One click to leave. The archive stays free either way.

We use your address to send the edition and nothing else. Confirm by email, leave in one click. How we handle it.