Circle Drawing · Tools

FREE BROWSER TOOLAI workflow

RAG Chunking Preview — Text Chunks and Overlap

Preview a fixed-size text splitter before preparing a retrieval dataset. Change the chunk size and overlap, inspect the actual boundaries, and copy one JSON object per line for a small indexing experiment.

Calculations run in your browser. No account or API key is needed.

Make it yours

An editable example is loaded below.

Your result

Source code points
339
Chunks
4
Repeated code points
72
Output / source
121.2%
{"id":1,"start":0,"end":120,"text":"Balcony garden notes\n\nMint grows well in its own container because its roots spread. Check the soil before watering; a s"}
{"id":2,"start":96,"end":216,"text":"oil before watering; a shaded pot may stay damp longer.\n\nBasil prefers a bright position. Pinch the top leaves regularly"}
{"id":3,"start":192,"end":312,"text":"the top leaves regularly and remove flowers when growing it for leaves. Keep a label with the planting date so observati"}
{"id":4,"start":288,"end":339,"text":"anting date so observations remain easy to compare."}

How to use this tool

  1. Paste one document, including the headings you want to preserve.

  2. Set the maximum chunk size and a smaller overlap in Unicode code points.

  3. Compare the chunks and repeated-character count, then copy the JSONL.

See what the boundary removes

A chunk can separate a heading from the sentence it describes. Overlap repeats nearby text so that adjacent chunks share some context, but increases the amount indexed. This preview uses fixed boundaries; it does not find topics, embed text or score retrieval quality. Test representative questions before choosing production settings.

Use reproducible offsets

Each record includes an ID, text, and zero-based start/end offsets; the end is exclusive. Whitespace is preserved. Offsets count Unicode code points, not bytes, JavaScript string indices or model tokens. Output is limited to 200 chunks and 200,000 text code points to keep the preview responsive.

Common questions

What chunk size is best?

There is no universal setting. Compare retrieval results on your documents; this tool measures boundaries and duplication only.

Can an emoji be split?

A Unicode surrogate pair stays together. Multi-code-point emoji and combining marks can still cross a boundary.