Tetrak OCR
GitHub ↗

Custodian of the forgotten — intelligent text recovery for the unruly archive, on your own machine.

Historical archives do not arrive as tidy, uniform PDFs. A real collection is a chaotic mix of linen postcards, dense multi-column trade journals, illustrated theatre handbills and hand-lettered vaudeville posters. One-size-fits-all OCR fails on this material because it assumes clean, high-contrast, linearly structured print.

Tetrak OCR is a small, free, local-first tool that turns those scans into full-text transcripts — text you can attach to the images in your catalogue or digital asset management system, so that collections become searchable by their contents, hard-to-read items gain a readable transcript, and screen readers have a text alternative to work with. It inspects each document, runs the viable local engines over it, scores their output without a reference transcript, and sets aside anything too poor to trust.

Tetrak OCR, scribe of the unruly archive. Four claims flow left to right. A magnifier picks over a heap of postcards, a theatre handbill, an open journal and torn newsprint: historical archives are inherently unruly, because postcards, handbills and journals lack the uniform structure standard OCR engines expect. Arrows fan into a rack of five geared bays labelled Tesseract, EasyOCR, PaddleOCR, Marker and Vision: routing beats picking a single engine, since the eligible local engines each read the file and the best transcript is kept. A server behind a padlocked shield: processing stays on your machine, with the Claude backend the one opt-in exception. The flow then splits into a folder marked CLEAN DATABASE and a bucket marked TRIAGE: a quality gate protects the archive, holding anything below the quality floor for a person to decide
The rack is an inventory of the local engines that ship, not a picture of what runs on any one file: auto-local fans out over the engines eligible for that file — four at most, two for a PDF. Vision is macOS-only and is not currently in that pool, though it measures as the strongest local engine on the corpus.

  • One interface, six engines

    Tesseract, EasyOCR, PaddleOCR, Marker, Apple Vision and Claude behind a single call. Swap engines with a flag; nothing else in your pipeline changes.

  • auto-local picks for you

    Runs every viable local engine on each file, scores the results with no reference text, and keeps the best one. Per file, not per batch. Name a single engine instead when throughput matters more.

  • Everything stays on your machine

    No cloud account, no API key, no material leaving the building. The one cloud path — a vision-model escalation for pages nothing local can read — is opt-in, per file, invoked by a person.

  • It tells you when it failed

    Anything below the quality floor goes to a triage/ queue with a manifest of what each engine produced. Bad transcripts do not enter your archive looking like good ones.

Two problems, one tool

No single OCR engine wins on mixed material. Measured over a corpus of Los Angeles stage and picture-palace ephemera (c. 1910–1945), five different engines win across nine documents. There is no setting you can choose once and be right:

Your archival materialBest local engineCharacter similarityWord recall
Printed captions and postcard reversestesseract-auto0.910.71
Low-contrast, textured illustration frontseasyocr0.870.71
Illustrated newspaper advertisingmarker0.750.88
Multi-page souvenir programmesmarker0.660.93
Dense small type on a playbillpaddle0.390.90
Hand-lettered display typepaddle0.130.39
Mixed collections, unattendedauto-local0.530.74

Choosing per file rather than per batch is worth a real margin: auto-local beats any single local engine on both metrics, entirely offline and free. The research has the full benchmark, the method and its honest caveats.

Some material cannot go to a cloud service. Donated collections often arrive with promises attached — donor agreements, privacy constraints, material about living people — and many institutions simply do not want collection scans leaving the building, least of all into services that might train on them. Tetrak OCR’s answer is structural rather than contractual: processing happens on your machine, nothing requires an account or a key, and the one path that sends a file to an API exists only as a deliberate, per-file decision by a person looking at the triage queue.

Who it is for

Curators of special collections, regional and county archives, local history societies, community archives, and museum collections rich in ephemera — the institutions with the most unruly material and the least appetite for cloud contracts. It runs on ordinary hardware; the reference environment is a laptop. If your aim is publishing out-of-copyright material openly, the full pipeline applies too, escalation included; if your collections carry restrictions, everything useful happens without any material leaving your machine.

Get it

pip install "tetrak-ocr[all]"
tetrak-ocr batch --backend auto-local

Installation covers the optional extras, system dependencies and a four-step quick start.

Where to go next

  • Get it running

    The quick start, optional backend extras, system dependencies and environment configuration.

  • Follow the tutorial

    Transcribe real archive images, batch a folder, and see exactly where OCR breaks — on the material that breaks it.

  • Read the research

    The corpus, the method, the full benchmark, which tool to reach for, and how each finding became a mechanism — including what the numbers do not establish, and one retracted finding.

  • Reference

    The command line and the Python API: the backend registry, accuracy scoring, and custom routing.