Turn screenshots into text, formulas, and tables you can use.
Offline OCR for papers, lecture notes, and everyday research.
Capture a region, review the result, and copy Markdown or LaTeX.
Get LocalTeX · Development · Report an issue
- Offline and private. Recognize locally without an account, API key, or cloud service.
- Images and handwriting. Extract text, formulas, and tables from screenshots, or write formulas on the drawing board.
- Ready to reuse. Copy Markdown or LaTeX, or open the result in Word.
- Review and revisit. Edit recognized content and find earlier results in a searchable local history.
Check Releases for available builds and release notes.
| Platform | Support |
|---|---|
| Windows x86_64 | Native desktop app, screen capture, global shortcuts, and system tray |
| Linux x86_64, X11 | Ubuntu 24.04 LTS (glibc 2.39); Ubuntu 22.04 is not supported |
| Linux Wayland | Not supported; use an X11 session |
| macOS | Outside the supported platforms |
Application packages include the OCR models for offline recognition.
Open an issue with your OS, the version shown in Settings → System, and steps to reproduce. For OCR issues, include an image you are comfortable sharing and the expected result.
LocalTeX is a single Rust package built with GPUI. Recognition uses local ONNX models: PP-DocLayoutV2 + UniRec-0.1B for images, and a separate stroke-based model for the drawing board.
- Development guide: prerequisites, build/run commands, model setup, checks, packaging, and project layout.
- Agent guide: project conventions and platform constraints.
LocalTeX is MIT licensed. Built with GPUI, ONNX Runtime, xcap, and RaTeX. Dependencies and model weights retain their respective licenses.