Installation¶
Requirements¶
- Windows, macOS 10.15 or newer, or Linux supported by Tauri 2.
- Node.js 20 or newer and npm.
- Rust through
rustup, with a toolchain compatible with the repository'srust-toolchain.toml(currently Rust 1.95 or newer). - The native build prerequisites listed by Tauri's prerequisites guide.
The application runs locally after it is built. A first build may need internet access to download npm packages and a matching PDFium archive.
Install and run¶
From the repository root:
The Vite frontend alone can be started with:
Use the Tauri development command when you need the desktop shell, Rust commands, encrypted storage, or native file dialogs.
PDFium¶
PDFium is the primary PDF text-extraction backend. The Rust build helper tries to download the target-matched archive into src-tauri/binaries. If that download is unavailable, the build continues with a warning so a manually supplied matching library can be used.
When a build reports that PDFium is missing, see troubleshooting. Do not copy a library built for another operating system or architecture.
OCR and models¶
The distributed build compiles the complete extraction feature set: PDFium,
Tesseract OCR, olmOCR-2 vision OCR, and Phi-4 repair support. Tesseract still
needs a native executable and language data. Settings → Ingestion can download
the official eng and por tessdata files into the app's managed models
folder; a reviewed native executable can be bundled at build time with
TESSERACT_BUNDLE_DIR, or installed using the linked platform guidance.
The embedded tiers provide explicit download buttons for their local model
assets and never download them during startup or ingestion.
Phi-4's recommended Q4_K_M file is about 2.5 GB. The complete olmOCR-2 7B snapshot is about 16 GB. Confirm that you have enough disk space before starting either download. These optional models are stored locally and remain under the app's model directory; they are not encrypted as part of the vault database, so protect the device and backups accordingly.
When building from source, the embedded tiers also require a native C++ toolchain, CMake, and LLVM/libclang for bindgen. The CI workflow installs these prerequisites on every supported platform.
The experimental olmOCR-2 and Phi-4 paths currently validate configuration and lifecycle state; they are not documented as a complete extraction guarantee. See implementation status.
Build a release bundle¶
Release builds are platform and architecture specific. Read releases and versioning before publishing an artifact.