ocrs
github.com
- Free plan
- Windows
- macOS
- Linux
- Web
- Self-hosted
What ocrs is
ocrs is an open-source Rust library and command-line tool that extracts text from images. It targets scanned documents, text-containing photos, and screenshots, with less preprocessing as a project goal. The CLI can write recognized text, return text and layout data as JSON, or create annotated images marking detected words and lines. An optional clipboard feature processes images copied from the system clipboard. Neural models are trained in PyTorch, exported to ONNX, and run with the RTen engine. On first use, required models download into `~/.cache/ocrs`. Developers can use the Rust library or the `ocrs-capi` C interface to create engines from model files or memory and retrieve text with line bounding boxes. The project is an early preview and may make more errors than commercial OCR engines. It currently recognizes the Latin alphabet, with broader language support planned. CLI installation requires Rust and Cargo.
Who it suits
ocrs suits developers building OCR into Rust or C applications and users comfortable with command-line tools. It is relevant for scanned documents, photos, and screenshots where Latin-alphabet recognition is sufficient.
Good to know, and what to watch
- Rust library and command-line interfaces
- JSON output includes text and layout
- Annotated image output marks detected lines
- Optional clipboard image processing
- WebAssembly build available
- Early preview with possible recognition errors
- Currently recognizes only the Latin alphabet
- CLI installation requires Rust and Cargo
- Models download on first run
The verdict
ocrs provides an open OCR stack with CLI, Rust, C, and WebAssembly options. Its preview status and current Latin-alphabet focus should guide production and language-support decisions.
What the maker publishes
- Automatic model download
- On first run, the CLI downloads required models automatically and stores them in `~/.cache/ocrs`. source
- C interface
- The `ocrs-capi` component exposes a C ABI that can create an engine from model files or memory buffers and return recognized text, including lines with bounding boxes. source
- C interface
- The project provides a C ABI with functions to recognize text and retrieve lines with bounding boxes. source
- CLI
- The CLI can extract text from image files and write text, JSON layout data, or annotated PNG output. source
- Clipboard
- The CLI supports reading an image from the system clipboard when installed with its clipboard feature. source
- Clipboard input
- An optional `clipboard` feature enables OCR on images from the system clipboard. source
- Image types
- The project aims to handle scanned documents, photos containing text, and screenshots with less preprocessing than older engines such as Tesseract. source
- Image types
- The project aims to handle scanned documents, photos containing text, and screenshots with little or no preprocessing. source
- Implementation
- Its neural network models are trained in PyTorch, exported to ONNX, and run using the RTen engine. source
- Installation
- The CLI can be installed with Cargo using `cargo install ocrs-cli --locked`; installing it requires Rust and Cargo. source
- JavaScript
- An alpha WebAssembly build is available for Node, browsers, and other WebAssembly runtimes. source
- Language support
- It currently recognizes the Latin alphabet, with broader language support described as planned. source
- Language support
- It currently recognizes the Latin alphabet only, with support for more languages planned. source
- Library API
- The project provides an OCR engine as a Rust library for use in Rust applications. source
- License
- The repository lists Apache-2.0 and MIT licenses. source
- License
- The repository lists Apache-2.0 and MIT licenses. source
- Maturity
- The project describes itself as an early preview and says it may make more errors than commercial OCR engines. source
- Model customization
- The C API can load custom models in ONNX format when built with the onnx feature. source
- Model data
- The project says it uses open and liberally licensed datasets, and links to a separate repository with model, dataset, and custom training details. source
- OCR engine
- Its neural network models are trained in PyTorch, exported to ONNX, and run using the RTen engine. source
- Output options
- The CLI can save extracted text, return text and layout information as JSON, or create an annotated image showing detected words and lines. source
- Platform support
- The project says it is intended to compile and run across platforms, including WebAssembly. source
- Project status
- The project describes itself as an early preview and cautions that it may make more errors than commercial OCR engines. source
- Purpose
- ocrs is a Rust library and command line tool for extracting text from images using OCR. source
Checked against the maker's pages on 9 Oct 2026.