GitHub radar
Turn PDFs and scanned documents into text on your own computer
Indra OCR is a model that turns photos and PDFs of documents into text and markdown, right on your computer. You can now read scans and handwriting without a GPU or a cloud subscription.
Indra OCR is a tiny OCR model (about 1M parameters) for recognizing text. It was built by Datalab, whose page on Hugging Face (a hub for AI models) is called datalab-to. The model runs on an ordinary CPU, no GPU needed. It turns PDF files and photos of documents into plain text. You can also get the result as markdown (a text format with headings and tables) or json (a structured format with block coordinates). The model keeps the reading order, headings and tables, reads both printed and handwritten text, works with tilted scans, and supports more than 80 languages. The code is released under the Apache 2.0 license — an open-source license. The model weights use a modified OpenRAIL-M license: personal use and use by startups earning under $20M a year are allowed. Using these weights to build a competing paid OCR service is not allowed.
When it helps
It's useful when you need to quickly turn scans, PDFs, or handwritten notes into text without expensive hardware or the cloud. The authors run their own paid document-recognition service, and the license forbids building a similar commercial service from these weights. So this model won't replace a paid subscription — it's a tool for your own personal tasks.
Pros
- Runs on a regular CPU, no GPU required
- Reads handwriting and more than 80 languages
- Beats Tesseract and PaddleOCR on complex documents
- Processes almost 44 pages per second on 32 cores
Cons
- Won't replace a paid subscription — the license forbids it
- Struggles with complex tables and handwritten text
How to set it up — step by step
- 1Get ready the PDF file or scanned document you want to recognize.
- 2Open your AI agent (Claude Code, Codex) in your project folder.
- 3Send the agent the text from the block below.
- 4Wait while the agent installs Indra OCR via pip and processes your file.
- 5Check the markdown result — compare headings and tables with the original.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install Indra OCR by the Datalab team for me: https://huggingface.co/datalab-to/indra-ocr — install it via pip, run one of my PDFs or scans through it, and show me the markdown result.
Any regular laptop or PC works — no GPU needed, the model is tiny and runs on the CPU.
Open on Hugging Face▌ More finds