Extract
Layout-aware parse for RAG and agents
Typed chunks and blocks with bbox and reading order. Feed retrieval and Claude/Cursor without losing the source line.
Parse output your pipeline can trust
01
Typed blocks
Headers, paragraphs, tables and figures come back labeled — not one text blob.
02
Reading order
Multi-column and multi-page order preserved, so chunks cite correctly.
03
Agent-ready
Same bbox evidence format as Extract. MCP serves chunks with citations.
Everything parse returns
Pick a capability to see what it does in Fieldry.
block 2 · tablebbox
block 4 · figuresummary
order1→2→3→4
Chunks & blocks
Layout segmentation with bbox per block.
- Header / text / table / figure types
- Bbox on every block
- Confidence per block
From upload to cited chunks
01
Upload any file
PDF/JPG/PNG up to 500 pages and 5GB. Scans and phone photos welcome.
02
Get ordered blocks
Layout model returns typed chunks with bbox in reading order.
03
Pipe to Extract or RAG
Reuse blocks via pipeline or feed your retriever with citations intact.
Related use cases
Parse returns layout chunks with bbox for RAG and agents. Extract returns your exact schema with validators. Run Extract on top of a Parse pipeline to reuse work.
Up to 500 pages per file and 5GB. Large files run async with webhooks; batch mode discounts bulk runs 50%.
No. Bbox and source text ship on every block — unlike competitors where citations are opt-in or disable chunking.
Yes. Scans, phone photos and handwriting are read with layout. Accuracy varies by language; low-confidence blocks are flagged, not hidden.
Feed agents grounded chunks
Layout-aware parse with bbox.
A 60-page file becomes ordered chunks with source boxes — RAG answers cite the exact line.