# Tesseract Tesseract is the default open-source (Apache 2.0) OCR engine — the free floor every document pipeline stands on. Started at HP in the 1980s, open-sourced and stewarded by Google from 2006, community-maintained since 2018. Version 4 replaced the classic character classifier with an LSTM engine; 100+ languages ship as trained models. ## Where it fits in 2026 Honest capability envelope: - **Strong**: clean scans of printed text, standard layouts, high-volume batch OCR at zero marginal cost - **Weak**: handwriting, complex multi-column layouts, tables, low-quality photos, anything where layout understanding matters as much as character recognition That envelope explains its role in modern stacks: [[liteparse]] bundles it as the default local OCR step, and routing-first tools like [[pdf-inspector]] exist precisely to decide *which pages* deserve more than Tesseract. The escalation path goes local Tesseract → vision-model OCR like [[Mistral OCR 4]] — paying per page only where the free floor fails. Usable as a CLI (`tesseract img out`), as `libtesseract`, and through wrappers in every language. ## References - [Tesseract on GitHub](https://github.com/tesseract-ocr/tesseract) ## Related - [[liteparse]] · [[pdf-inspector]] — pipelines that route around its limits - [[Mistral OCR 4]] — the paid escalation tier - [[MarkItDown]]