An SEO tool to help you test different chunking models on your content and see how the relevance score changes for different keywords.
Everything runs locally in your browser — nothing is sent to a server. The first analysis downloads a small open-source embedding model (~90MB, cached after that).
The tool
Try it a few different ways: run the same text through each strategy, tighten a rambling paragraph and rerun it, or add a query and watch the match score move as you edit. Nothing here is saved — reload and start over as many times as you want.
Add keywords or questions to see which chunks would be retrieved for each one.
Strategy guide
The full text is treated as a single block. Useful as a baseline to see how retrieval behaves without splitting content at all.
Sentences are grouped by embedding similarity, so each chunk stays topically coherent. This is closest to how modern RAG pipelines chunk content.
Based on semchunk ↗Text is split every N words or characters, regardless of meaning. Simple and predictable, but can cut sentences or ideas in half.
Splits on paragraphs, headings, and lists — the structure a human editor already gave the content. Good for well-formatted pages.