Chunk preview for RAG
See exactly how your document splits before you index it — heading-aware boundaries, adjustable size and overlap, and JSONL export with metadata.
Retrieval quality is decided at chunk time, not query time. Drop a document, open the Chunks tab, and see the real boundaries: heading-aware by default, with the heading path and page number carried into every chunk. Drag the size (100–2,000 tokens) and overlap sliders and watch the plan change. Export JSONL and your pipeline ingests it directly — no glue code.
Frequently asked questions
What's a good chunk size?+
A common working range is 600–1,000 tokens with 50–150 overlap. Heading-aware boundaries tend to beat fixed-size for Markdown because they keep whole sections together.
What's in the JSONL?+
One object per chunk with the text plus metadata: source, title, chunk index, heading path, page number and token count.
Is it heading-aware?+
Yes by default. Chunks prefer to break at headings so your retriever returns whole sections instead of sentence fragments. You can toggle it off.
Does my document upload?+
No — chunking runs in your browser like everything else.