Install & first launch

Piixie is a local app. Install it, launch it, and after a one-time model download you can anonymize documents without a network connection.

System requirements

• macOS (Apple Silicon or Intel), Windows, or Linux
• Around 6 GB of free disk space for the default model
• 8 GB of RAM or more. Inference runs on your CPU/GPU through llama.cpp, so more memory and a faster machine mean faster anonymization.

If your machine is underpowered, you can skip local inference entirely and point Piixie at a remote endpoint or a Piixie server on your network.

Install

Packaged builds for macOS and Windows are on the download page.

Developers can run from source:

git clone https://github.com/rodrigolive/piixie.git
cd piixie
make run

First launch: the model download

On first launch Piixie downloads its default local model, Gemma 4 E4B (~5 GB), plus a small vision projector file. A progress bar tracks the download; the files are cached under your user data directory, so this happens once.

Platform	Model cache location
macOS	~/Library/Application Support/Piixie/models/
Windows	%APPDATA%\Piixie\models\
Linux	~/.config/Piixie/models/

After the download finishes, Piixie starts a llama-server process in the background and the status indicator turns ready. From that point on, anonymization runs entirely on your machine.

A larger model, Gemma 4 12B (~7.1 GB), is available from the model selector if you want higher quality and your hardware can handle it. See Local models.

Vision included, no OCR step

Both bundled models support vision. Piixie analyzes rendered pages and embedded images directly, so scanned PDFs and image-backed documents work without a separate OCR pass.

Next step

Anonymize your first document.