Document prep for RAG pipelines
Convert, chunk and export in one place — heading-aware boundaries, per-chunk metadata, JSONL your retriever ingests directly.
RAG quality is decided before the query ever runs. MarkPrep converts your source documents to clean Markdown, strips the page furniture that poisons embeddings, and chunks heading-aware so retrieval returns whole sections instead of fragments. Every chunk carries its heading path and page number, and JSONL export drops straight into your pipeline. We've preset the chunk size to a RAG-friendly default here — adjust it in the Chunks tab.
- Heading-aware chunking with visible boundaries
- Per-chunk metadata: heading path, page, token count
- JSONL export — no glue code
- Clean-up removes repeated headers/footers before they hit your embeddings
Frequently asked questions
What chunk size should I use for RAG?+
A common working range is 512–1,000 tokens with 50–150 overlap. Heading-aware boundaries usually beat fixed-size for Markdown. We default this page to 512/64 — tune it live.
What metadata is in the JSONL?+
Each line has the chunk text plus source, title, chunk index, heading path, page number and token count.
Is anything uploaded?+
No — conversion and chunking run in your browser.