arcships/light-ocr

light-ocr is a tool that reads text out of images, running entirely on your own computer with no internet connection needed. You give it a JPEG, PNG, or raw image data, and it hands back each line of text it found, in reading order, along with a confidence score and the exact coordinates of where that text sits in the image. For Node.js developers, installing the npm package gives you everything needed in one step, including the OCR model itself and a prebuilt component for macOS, Linux, and Windows. There is also a command line tool that works right after installing, letting you recognize text, detect text regions only, or check engine info from a terminal. An Agent Skill is also included so AI agents that can run local commands can call light-ocr directly, with guidance on when to use OCR instead of a general-purpose AI model that also reads images. One of the main selling points is speed through hardware acceleration. On Apple Silicon Macs it automatically uses Apple's Core ML technology, and on Linux and Windows it uses WebGPU, falling back to the regular processor if neither is available. According to the project's own measurements, this hardware acceleration made recognition noticeably faster and used far less processing time compared to running on the CPU alone, though results vary by device and workload. For C++ projects, the library can be built from source and linked directly, with the same ability to process image pixel data. All processing and results stay local to your machine: nothing is uploaded anywhere. The project is available under the Apache License 2.0.