Skip to content

ReadAnyDoc documentation

ReadAnyDoc is the open-source alternative to Reducto. It parses and extracts PDFs, scans, photos and Office files with a vision model (Claude Haiku 5.5 by default, or any model you choose) and returns structured data where every value carries a citation back to the place on the page it was read from.

You can use it two ways:

  • As a Python library. pip install "readanydoc[engine]", then readanydoc.parse(...) and readanydoc.extract(...). No server, nothing to deploy.
  • As a server with Reducto's API. Run readanydoc serve or the Docker image and call it over HTTP from any language, or point Reducto's own SDK at it.

Either way it runs on your machine and calls the model provider with your own key. Nothing is sent to us.

  • Quickstart (Python)

    Install, parse your first document and extract into your own pydantic class in five minutes.

    Start here

  • Migrate from Reducto

    Keep your code and Reducto's SDK; change the base URL. Route-by-route compatibility included.

    Switch in one URL

  • Choose your models

    Claude, OpenAI or a self-hosted model, with the effort and every provider option you need.

    Choosing models

  • Benchmarks

    Measured against Reducto on five public benchmarks, about 4,250 pages and 10 long documents.

    See the results

What it does

Function What you get Library Server
Parse Typed blocks (text, tables, form fields, figures, headers) with positions, Markdown, chunks for search readanydoc.parse POST /parse
Extract Your schema filled in, with a citation for every value readanydoc.extract POST /extract
Split Where each document starts and ends inside one combined PDF POST /split
Classify A document type label, with the text that supports it POST /classify
Edit PDF forms filled in, including flat and scanned ones POST /edit
Jobs and webhooks Every call as a background job, with a callback when it finishes *_async, GET /job

Two modes

Mode What runs Cost per page
engine (default) The model reads each page; ReadAnyDoc adds layout, form fields, table checks, zoomed re-reads of scans and word-level citations. Handwritten pages go to Claude Sonnet 5.5. about $0.003 to $0.005
model-only One model read per page and nothing else about $0.001 to $0.002

Read more in Modes. Reducto charges $0.010 a page for comparison.

Where to go next