

Velum Privacy
Masks personal data in text and files before it reaches ChatGPT, Claude or any AI, then restores it in the reply, running entirely on your own device.
Cost / License
- Freemium (Subscription)
- Proprietary
Platforms
- Mac
- Google Chrome
- Jira
- Brave
- Confluence
- Mozilla Firefox
Features
Properties
- Privacy focused
- Local-First
Features
- Dark Mode
- Works Offline
- AES-256 Encryption
- Extensible by Plugins/Extensions
- No registration required
- Ad-free
- No Tracking
- Offline
Velum Privacy News & Activities
Recent activities
mhrsntrk added Velum Privacy as alternative to Minimal eReader
mhrsntrk added Velum Privacy as alternative to LEKTRA
mhrsntrk added Velum Privacy as alternative to Orion Viewer- mhrsntrk updated Velum Privacy
mhrsntrk added Velum Privacy as alternative to Secure PDF Viewer
mhrsntrk added Velum Privacy as alternative to Document Viewer- mhrsntrk added Privacy Tools as a feature to Velum Privacy
- mhrsntrk added Velum Privacy
- mhrsntrk rated Velum Privacy
- mhrsntrk liked Velum Privacy
Velum Privacy information
What is Velum Privacy?
Velum Privacy replaces personal data with reversible tokens before you send text or files to an AI assistant, and puts the real values back when the answer comes home. Both the masking and the restore run on your own machine.
The round trip is the point. One-way redaction protects the data but breaks the answer, because the model can no longer tell you which person or which invoice it is talking about. Velum keeps a local mapping, so a masked name comes back as the real name in the reply and the assistant still returns something useful.
It ships on two surfaces that share one detection engine. The browser extension is a popup: paste text in, mask it, copy the result into any web chat. It sits alongside ChatGPT, Claude, Gemini, Perplexity, Copilot and Grok because it works through the clipboard rather than hooking into any one site. The desktop app adds documents and whole folders, a visible mapping ledger, an audit log, and an MCP server so coding agents read masked files instead of the originals.
Detection combines deterministic pattern matching, checksum validation for structured identifiers such as IBANs and national ID numbers, and a local NER model for the things patterns cannot catch, such as a name that is only identifiable from context. It covers nine language variants across Spanish, Catalan, English, German, French, Italian and Turkish. Accuracy is published rather than asserted, including the failure rates on messy real-world text.
The privacy claims are meant to be checked rather than believed. The browser extension declares no host permissions at all, which the browser itself shows you at install time, so it is structurally incapable of contacting a website. The desktop app names every outbound call it makes in a Protection panel, none of which carries document text, and hands you the loopback address and the shell command to confirm that yourself. Turn the network off and the built-in self-test still masks.
Licensing is deliberately plain. Text masking in the browser extension is free, with no key and no time limit. File masking and the in-page button need a licence. The desktop app runs a seven-day trial with no account and no card, then needs the same licence, which is an offline key verified on the machine rather than a server check. Unmasking is never gated, so anything you have already masked can always be recovered even if a licence lapses.
The detection engine is source-available under PolyForm Noncommercial. That is not an OSI open-source licence and the packages are not published yet, so it should not be described as open source. The applications themselves are proprietary.





