featured · github
docTR: Open-Source OCR That Doesn't Lock You Into a Cloud Bill
A deep-learning document recognition library that runs locally, handles layout parsing, and cuts the vendor dependency out of your document automation workflows.
mindee/doctr ↗docTR is OCR (optical character recognition) software that reads text and structure from scanned documents and PDFs. Unlike cloud APIs from Google or AWS, you run it on your own hardware or servers, so there's no per-page fee or data leaving your system. It handles both the text extraction and the layout parsing (figuring out where fields are on a page), which means you can automate invoice processing, form data entry, or contract review without manual intervention. The trade-off: you manage the infrastructure yourself. If you're shipping a product that needs document intake at scale, this saves licensing costs. If you're building automation for a client, it removes vendor lock-in and keeps their data private.
Share kit
docTR: Self-hosted OCR, no API bills
Deep-learning text recognition that runs locally. Handles documents and layout parsing. Good fit if you're automating document workflows at scale and want to avoid per-page cloud API costs or vendor lock-in. GitHub link in the report.
docTR is a solid pick if your product swallows documents. Deep learning OCR plus layout parsing, runs on your own hardware, no per-page API toll. Open source. Worth a test if you're tired of vendor lock-in or building something that needs text recognition at volume.
Building document automation. docTR is an open-source OCR library that extracts text and parses layout from scanned documents and PDFs. Runs locally, no cloud API costs, no vendor lock-in. Clean fit for automation workflows at scale.
Just integrated docTR into a document automation workflow and stopped paying cloud OCR vendors for every page. This is a local-first, production-grade alternative that handles text recognition and layout parsing without the per-request tax. What sold me: built on deep learning, no black box, runs on CPU or GPU, and the library is genuinely maintained. You get structured output (bounding boxes, confidence scores, page geometry) which means downstream automation actually works instead of garbage-in-garbage-out. If you're building anything with document intake (contracts, invoices, forms), this cuts both latency and cost. Worth the swap from Tesseract or API dependency. Github: https://github.com/mindee/doctr
docTR is a legitimately good OCR library. Local inference, structured output, no vendor lock-in. If you're building doc automation and still calling cloud APIs for every page, you're paying rent on someone else's model. https://github.com/mindee/doctr