ocrqueen 0.4.0

Official Python SDK for the OCRQueen document extraction API

ocrqueen-python

Official Python SDK for the OCRQueen document and image extraction API.

🚧 Status: Pre-release. APIs and surface area will change before v1.0.0.

Installation

pip install ocrqueen

Requires Python 3.10 or newer.

Supported formats

Category	Formats
Documents	PDF
Presentations	PPTX, PPT (PowerPoint)
Images	PNG, JPEG, WebP, HEIC / HEIF (iPhone photos)

The API returns structured JSON + Markdown for every supported type — text, tables, images, and (with extraction_profile="advanced") diagram graph extraction and image alt-text.

Quickstart

from ocrqueen import OCRQueen

client = OCRQueen(api_key="pk_...")

with open("paper.pdf", "rb") as f:
    job = client.extract.create(file=f)

result = client.jobs.wait(job)
print(result.result["markdown"])

Get an API key from dashboard.ocrqueen.com.

Other file types

# Slide decks — speaker notes are preserved
job = client.extract.create(file=open("pitch.pptx", "rb"))

# iPhone photos — HEIC handled natively, no conversion needed
job = client.extract.create(file=open("receipt.heic", "rb"))

# Scanned document images
job = client.extract.create(file=open("invoice.png", "rb"))

# Deeper extraction profile — diagrams, image alt-text, OCR on
# embedded text
job = client.extract.create(
    file=open("paper.pdf", "rb"),
    profile="advanced",
)

Patent extraction (domain="patent")

Route a PPTX or PDF through the patent-specific pipeline: region classification (cover / abstract / drawings / claims / references), Gemini cover parser, LibreOffice rasterisation for EMF/WMF figures, cross-figure numeral resolution, and an honest per-stage faithfulness_score. Billed flat at $0.05/page regardless of profile.

job = client.extract.create(
    file=open("invention-disclosure.pptx", "rb"),
    options={"domain": "patent"},
)
result = client.jobs.wait(job).result        # response shape changes — discriminator is `domain`
patent = result                              # full PatentExtractionResponse
print(patent["source"]["input_kind"])        # "invention_disclosure" | "published_patent" | "unknown"
print(patent["extraction"]["faithfulness_score"])

# Figures carry a stable proxy URL — never expires until the underlying
# object is purged by your retention window. fetch_image() handles the
# 302 → signed-storage dance for you and returns raw bytes.
for fig in patent["drawings"]:
    bytes_ = client.jobs.fetch_image(fig["image_url"])
    open(f"{fig['figure_number'].replace(' ', '_')}.png", "wb").write(bytes_)

The same fetch_image() helper works for general-domain ImageBlock URLs (pages[].blocks[].url) — useful for snapshotting all figures from a job into your own pipeline.

Documentation

• Full API reference: https://ocrqueen.com/docs
• Python SDK guide: https://ocrqueen.com/docs/sdks/python
• Data retention & deletion: https://ocrqueen.com/docs/data-retention

License

MIT — see LICENSE.