Python library

import readanydoc

The library runs the same pipeline as the server, in your process. It covers parse and extract; split, classify and edit are server features.

readanydoc.parse

readanydoc.parse(
    source,                      # path (str or Path) or bytes
    *,
    mode="engine",               # "engine" | "model-only"
    model="claude-haiku-5-5",    # page reader
    effort="high",
    model_options=None,          # dict added to the provider request
    handwriting_model="claude-sonnet-5-5",   # None = read handwriting like any page
    handwriting_effort=None,
    handwriting_options=None,
    pages=None,                  # list of 1-based page numbers
    citations=False,             # word and cell positions
    filename=None,               # needed with bytes
    options=None,                # any Reducto parse option (retrieval, formatting, settings, ...)
    timeout=None,                # seconds
) -> ParseResult

readanydoc.extract

readanydoc.extract(
    source,
    schema,                      # pydantic class or JSON Schema dict
    *,
    instructions=None,           # system prompt for the extraction model
    citations=True,
    array_extract=False,
    mode="engine",
    model="claude-haiku-5-5",
    effort="high",
    model_options=None,
    extract_model=None,          # default: model
    extract_effort=None,         # default: effort
    extract_options=None,
    handwriting_model="claude-sonnet-5-5",
    handwriting_effort=None,
    handwriting_options=None,
    pages=None,
    filename=None,
    settings=None,               # any further Reducto extract setting
    timeout=None,
) -> ExtractResult

Result types

All are pydantic models.

Type	Fields
ParseResult	blocks: list[Block], pages: int, usage: Usage, warnings: list[str], raw: dict; properties markdown, tables
ExtractResult	data (your class instance, or a dict if it did not validate), citations: dict[str, list[Citation]], usage, warnings, raw
Block	type, content, bbox: BoundingBox \| None, confidence: "high" \| "low" \| None
Citation	content, bbox, confidence
BoundingBox	left, top, width, height (fractions of the page), page (from 1)
Usage	calls, input_tokens, output_tokens, cost_usd (None when a model has no known price), seconds, by_model

Errors

Exception	When
readanydoc.pipeline.PipelineTimeout	the document took longer than timeout
readanydoc.modelspec.ModelSpecError	an unknown model name or a reserved request option
FileNotFoundError	the path does not exist
provider errors	wrong key, rate limits after retries, model not available