bibiocr: Offline Cross-Platform OCR Desktop Tool with Rust/C++ and Local AI Models

Project Overview

bibiocr is an offline OCR desktop application developed in Rust and C++ with support for Windows, macOS, and Linux. The project is hosted at https://github.com/bibiparrot/bibiocr. It can process single images or entire folders, read screenshots directly from the clipboard, and export recognition results as structured Markdown or Word documents.

Key Design Points

1. CJK Text Friendly

The application bundles the Noto Sans CJK font to guarantee consistent rendering of Chinese, Japanese, and Korean characters across all supported macOS versions, eliminating garbled or missing glyphs.

2. Thoughtful Configuration Panel

The configuration panel opens the active bibiocr.toml file directly, validates the configuration before saving, and displays the actual path of the user configuration file — a convenience for users who prefer manual parameter tuning.

3. Models Not Pre-installed; First-Run Selection

Models and third-party runtimes are not packaged in the installer. On first launch, a "Download & Dependency Configuration" interface appears, allowing online download or selection of local files. The configuration reveals a hardcore technical stack: PaddleOCR-VL GGUF models for recognition, ONNX models for layout analysis, llama-server for LLM integration, and pdfium for PDF handling.

4. Batch Conversion with Office Document Support

In "Batch Processing" mode, users select input and output folders. Common office documents are converted via AnyDoc, while images and scanned PDF pages are processed by bibiocr itself. Each input file produces both a .md and a .docx output.

5. Download Experience Optimized for Chinese Users

Model files are fetched through the official hf-hub Rust client. The UI lets users choose between huggingface.co and the domestic mirror hf-mirror.com, with support for proxy, retry, and resume. GitHub Release assets are downloaded via the built-in bibiget library, which provides eight concurrent connections, resume capability, proxy support, GUI progress display, and a customizable GitHub accelerator URL.

Language Support

The interface is available in six languages: English, Chinese, French, Russian, Korean, and Japanese.

Project Status

As of writing, the repository has 64 stars and 7 forks — a niche but well-polished open-source project.

Conclusion

bibiocr focuses on "offline usability" and "cross-platform consistent experience." Its attention to detail — font rendering, configuration validation, and download acceleration — makes it a strong candidate for privacy-sensitive users who prefer not to upload images to cloud-based OCR services.