Skip to content

Piixie documentation

Anonymize documents on your own machine. A local language model finds the PII, Piixie rewrites the file, and the original never leaves your computer.

Piixie is a desktop app for macOS, Windows, and Linux that removes personally identifiable information from documents before you share them. Drop in a PDF, Word file, Markdown or plain text document, pick an anonymization mode, and Piixie writes a safe copy next to the original. Detection runs on a local Gemma model through llama.cpp, so the raw document stays on your machine unless you explicitly configure a remote endpoint.

Install and first launch

Download the app, let it fetch the local model, and anonymize your first file. Get started

The Anonymization Editor

Review what the model found, fix it by hand, draw redactions, and regenerate — in a side-by-side window. Open the editor

Anonymization modes

Redaction, replacement tokens, synthetic data, LLM generation, or your own JavaScript. Compare modes

Dictionaries & templates

Keep the same fake identities across documents, and reuse your edits on similar files. Dictionaries · Templates

Deanonymization

Anonymize, use an external tool safely, then put the real values back. The round trip

Recipes

End-to-end walkthroughs: CVs, insurance claims, medical records, team data, safe LLM use. Browse recipes

Privacy and data storage

What stays on disk, where it lives, and what (if anything) crosses the network. Read the details

Models, local to cloud

Run locally, on a shared GPU server, or via Anthropic/OpenAI — with optional reasoning. Choosing a model