Privacy and Offline

Privacy and Offline

Short Answer

The current project documentation states that AI-assisted features are intended to run locally on the user's machine.

That means:

• prompts and images are not uploaded to a hosted cloud inference service for semantic search, similarity, captions, or prompt alignment
• internet access is mainly required for first-time model download
• once models are cached locally, AI features can work offline

What Uses The Network

The main network-facing part of the AI workflow is model download from HuggingFace on first use.

After the models are cached locally, the normal AI workflows can operate offline.

Token Clarification

The docs now clearly separate two token concepts:

• HuggingFace token: optional, helps with model download access and rate limits
• Majoor API token: protects remote write access to the local Majoor backend

These are different concerns. The Majoor API token is not described as a hosted AI inference token.

Why This Page Matters

This distinction is easy to miss if a user only reads the README or sees the word token in settings. The dedicated privacy/offline guide is one of the most useful additions to the documentation set and deserves a first-class wiki page.

Canonical Docs

• Privacy and Offline Guide
• AI Features Guide
• Security Environment Variables
• Settings Configuration