Hide personal details before you send them to AI.

Sether Shield is an extension for Chrome on a laptop or desktop, so it protects prompts on your computer. On your phone you can try the browser demo, which runs on the page with sample text and makes no AI request, and send yourself the install link for later.

No. It finds the types it supports, and results depend on the format of the text and your settings. Review the edited prompt before you send it. Pro adds your own watchlist terms and rules for details specific to your work.

The full npm and PyPI libraries stay free and MIT licensed, with no Sether usage quota. Official tools keep scanning, masking, token replacement and restoration of all known mappings in Free. New licensed releases add Pro realistic decoys and custom watchlists, plus custom rules in Shield: one user across Shield and the editor, $5 for a month or $48 for a year, with manual renewal. The opt-in seven-day trial needs no card and never charges automatically. Install a compatible pilot build before starting; current store builds may retain legacy free features.

Many teams use the term PII, short for personally identifiable information. If the acronym does not resonate with yours, think personal data: names, emails, phone numbers, payment details, and anything that could identify a real person. Sether tokenizes those values before they reach the model.

Great if you have a Python team and need batch processing. Sether is streaming-native and TypeScript-first, built for the JS/TS stack the AI dev community actually uses, with a Python port that uses the same engine and token format. Often complementary, not directly competitive.

Decoy mode replaces supported sensitive values with realistic stand-ins, such as example.com emails and documented test ranges where available. A matching local vault can restore recognized replacements; a changed or unknown replacement may not restore. The full npm and PyPI libraries include decoy mode free under MIT. In new official Shield and editor releases, realistic decoys are part of Pro alongside custom watchlists; custom rules are Shield-only and known-mapping restore remains Free. Legacy published builds may retain their existing free features.

You can, and we won't pretend basic detection is hard. The hard parts are the ones you would own forever: values split across streaming chunks, restoring tokens safely into JSON and tool calls, keeping 200+ patterns safe from ReDoS, and keeping up with new API key formats. That is what the library is.

With the npm package, streams are processed inside your Node process. The default MemoryVault keeps token→value mappings in RAM for that process; implement the Vault interface when you need shared or durable storage. This site's server playground runs Sether only to handle your pasted sample in that request. It is not a substitute for your own security review.

Most teams wire Readable streams through redact() and restore() in an afternoon. Drop-in Express / fetch / OpenAI SDK helpers ship in the npm package so you can skip some boilerplate.

Worst case, you fork the OSS core. It is MIT licensed and the npm package is yours forever. Best case, we keep shipping. Either way, no vendor lock-in by design.

No. Sether does not currently hold a SOC 2 report. The core libraries are MIT licensed and run inside your application; inspect and assess them for your own requirements. A Pro subscription is a feature license, not a compliance certification.

Yes. The OSS primitives are plain text transforms. You can tokenise emails, phones, IBANs, IPs, and similar patterns in listings or logs, then restore only where your app authorises it. The opt-in identity pack (0.3.0) also redacts label-anchored names, dates of birth, passport numbers, and addresses; names in plain sentences are handled by the sether-ner add-on on npm. Today you implement policy in your own code.