Configure Structural Nodes and Chunking
This stage converts parsed Markdown into semantic nodes and retrieval-ready chunks.
| Setting | Effect |
|---|---|
| Max Chunk Tokens | Maximum chunk size. Smaller chunks are more focused; larger chunks preserve more surrounding context. |
| Chunk Overlap | Context repeated between neighbouring chunks to reduce boundary loss. |
Start with the displayed defaults unless testing shows a retrieval problem. After running the stage, use View / Edit Chunk JSON or Browse Nodes.
Browse Nodes can filter by All, Text, Table, or Summary, filter by page number, search content or section, and expand a node to inspect its text, section path, kind, page, and token count.
If you rerun chunking, review dependent table summaries and enrichment and re-embed before relying on the new output.
Still stuck? See Get Help with Support or reach your workspace administrator.