Unstructured is a broad ingestion pipeline for RAG. If you mainly need PDFs turned into clean, structured data you can trust, this is the focused option.
Unstructured is built for feeding many document types into LLM and RAG pipelines, with chunking and a large set of source connectors. There's an open-source core you can self-host, which suits broad ingestion workflows.
pdftojson does the narrower job well: it turns PDFs into clean JSON, Markdown, CSV, XML or text, with Markdown that keeps headings for chunking, and a correctness check for the financial documents where a wrong value actually matters.
Unstructured is designed around getting many file types into LLM and RAG pipelines, with chunking and a wide set of connectors.
There's an open-source library you can self-host, which is appealing for privacy-sensitive or high-volume workloads.
| pdftojson | Unstructured | |
|---|---|---|
| Focus | PDF to structured data you can trust | Multi-format ingestion for RAG (many file types, connectors) |
| Output formats | JSON, Markdown, CSV, XML, Text | Elements/JSON for chunking |
| Free in-browser tool | Yes | No (library or API) |
| Bank-statement reconciliation | Yes | No |
| Setup | One API call | Library or API plus connectors |
| Pricing | Flat $29/mo | Open-source / usage-based API |
If you need a broad pipeline that ingests every file type and wires into a dozen sources, Unstructured is built for that. pdftojson is for the case where the document is a PDF and what you want back is clean, structured data without standing up a pipeline.
And for financial documents it adds a check an ingestion pipeline doesn't: the bank-statement endpoint reconciles the transactions against the opening and closing balances and flags any that don't add up.
Yes. The Markdown output keeps headings and structure, which chunks better than raw text, and the paragraph JSON shape is a good starting point for embeddings.
When you're ingesting many file types (not just PDFs) into a RAG pipeline and want its connectors and chunking. pdftojson is focused on clean PDF extraction with a correctness check.
No, it's a hosted API plus a free browser tool. If self-hosting is a hard requirement, Unstructured's open-source core fits that; pdftojson trades that for zero setup.
A flat $29/mo Developer plan with a free tier, billed per page processed.