Reducto

Parse turns PDFs, scans, spreadsheets, and slides into structured, citation-ready JSON. It's the document extraction engine everything else builds on. It handles the hard cases: complex tables, charts, equations, handwriting, and multilingual OCR. Each value comes back with a reference to where it sat on the page, so downstream checks don't have to trust the model blindly.

Extract returns typed JSON that matches a schema you define, which is where most of the real document extraction happens. You ask for the fields you want, and Reducto pulls them out with citations and bounding boxes attached. There's a Deep Extract mode for documents where layout is inconsistent enough that a single pass isn't reliable.

Split cuts a single document into meaningful sections or repeating sub-documents, which helps when one upload contains twenty invoices instead of one. Document classification sorts each file against a plain-language taxonomy, so you can describe categories in normal words and let Reducto route it.

Edit turns extracted data back into finished PDF and DOCX files. Instead of returning a table for someone to re-type, it writes the values into your template. Pre-filled pages cost less than blank ones, so repeat forms stay cheap.

Studio is a no-code environment for building, testing, and deploying document pipelines. You chain Parse, Extract, Split, and Classify steps together and preview the output before it ships. Standard accounts include up to 5 seats; higher tiers lift that cap.

Reducto ships Python and Node.js SDKs, a CLI you install with pip install reducto-cli, and a hosted MCP server that lets AI agents process documents inside their own reasoning loop. There's also an OpenAPI 3.1 spec, so you can generate clients if the official SDKs don't fit.