Text chunker
Compare fixed, sentence and Markdown splitting on your own text, with every chunk drawn to scale.
/chunk
Free · no account · nothing uploaded
RAG playground: retrieval you can actually inspect
Load a document, ask it a question, and watch a real BM25 ranking do its work — with the chunk map, the term statistics, and the exact arithmetic behind every score on screen.
RAG playground
Say it the way you would say it to a person. Your words are matched to one of 18 tools by readable rules, and the tool runs right here in your browser. No upload, no account.
This tool runs in your browser and needs JavaScript turned on.
esc
How to use the RAG playground
- Add a document: drop a .txt, .md, .csv or .html file into “Your files”, or paste the text.
- Ask it a question in the box — “what is the refund policy?” — or search it for a phrase.
- Read the best passages, each cited to the chunk it came from, with its BM25 score.
- Expand a result to see which of your words contributed to its score, and by how much.
Things you can type: what is the refund policy? search it for security reviews
Retrieval is not magic, and you should be able to see that
Vector search has made retrieval feel like a black box: an embedding goes in, neighbours come out, and when the wrong passage is returned there is nothing to look at. BM25 is the opposite. It is a scoring function with visible parts — term frequency, inverse document frequency, and a length normalisation — and every result can be explained down to the contribution of a single word.
That makes it the right thing to learn on. Once you can see why a passage won, the behaviour of a hybrid or dense retriever stops being mysterious and becomes a question about which signal you traded away.
What you get for each query
- A chunk map of the whole document, with the matching chunks highlighted in place.
- The frequent terms of the document and of your query, so you can see which words are actually doing the work.
- Ranked excerpts, each cited back to its chunk rather than paraphrased.
- The score breakdown per result: which terms contributed, and how much each one added.
Deterministic, and no AI involved
There is no model here and no embedding service. The same document and the same query produce the same ranking every time, which is what makes the explanation worth reading.
It also means the tool costs nothing to run and needs no key: the arithmetic happens on your machine, in the tab.
From playground to research workspace
A playground shows you the mechanism on one document. The problem that follows is scale: hundreds of sources, mixed formats, and answers that need to cite the original page months later.
That is what FlexContext is being built for — a persistent library where retrieval runs over your own sources and every claim can be opened at the exact passage it came from.
Questions
RAG playground: common questions
What is BM25?
A ranking function that scores a passage against a query using how often the query terms appear in it, how rare those terms are across the corpus, and how long the passage is. It is the default lexical ranker in Elasticsearch, Lucene and OpenSearch, and it remains a strong baseline against dense retrieval.
Does this use embeddings or an AI model?
Neither. The ranking is pure lexical arithmetic, computed locally and deterministically, which is precisely why every score can be shown and checked.
Can I search a PDF with it?
Not yet. It accepts pasted text and text-based files — Markdown, TXT, CSV, HTML. PDF extraction needs a vendored parser, and adding one honestly is still on the list.