One detail, or the whole prompt.
Accept or dismiss individual findings, choose a suggested decoy with Pro in new releases, or use Scrub all. Keep the final decision in your hands.
Sether Shield / Chrome extensionKeep the context.
Change the details.
Your AI can help write the email. It doesn't need your actual email address.
Shield flags supported personal data and secrets as you type. Free masks and tokens keep the round trip useful. Optional Pro adds realistic decoys and custom controls in new official releases.
- Free essentials
- Optional Pro
- Local text processing
The current store build may retain legacy free features. See the Pro pilot and compatible builds. The demo shows both Free and Pro workflows; it does not start a trial.
Sether ShieldPRIVATE DETAILS. USEFUL AI.01 / DETECTSpot the private details.
A second look, before you send.
In your draft
maya@northstar.exampleReview the match
Email address foundDetection leaves you in control.
AI chat · illustrative demoDEMO
YOUR PROMPTShield enabled
Draft a reply for my client Maya Chen. Email: maya@northstar.example. Phone: +1 202-555-0148.
Sether Shield3 foundNameEmailPhone
TRY THE EXPERIENCE
A second look, before you send.
Review supported personal details in the prompt. This example has a name, email and phone number.
Explore the controls, or watch the guided tour.
Fictional data. No AI requests. Reply restore supports ChatGPT, Claude and Gemini; compatibility depends on the site. Explore all features
In the AI tools you already use
- ChatGPT
- Claude
- Gemini
- Microsoft Copilot
- Perplexity
- DeepSeek
- Grok
- Mistral Le Chat
- Poe
- Hugging Face
- Character.AI
Prompt protection on these default sites. Add other compatible sites in the popup.
A more considered workflow
From the first paste
to the final reply.
Small choices at the right moments. Useful context for your AI, more control over the details you share.
01 Before you send
A useful prompt.
A fresh set of details.
Mask sensitive values or replace them with realistic stand-ins. Choose a decoy for an individual finding, or scrub the whole prompt at once.
Original
maya.chen@example.comMask
m***@e***.comDecoy
lena.fischer@example.comPro in new licensed releases. Fictional example data; suggestions vary.
Decoy mode
A local swap, before you send
Detected email
maya.chen@example.comChosen decoy
lena.fischer@example.comMatched details change while the surrounding prompt stays readable. Suggestions vary; this is fictional example data.
Input or known replacementResult you chooseIllustrative example
02 When the reply arrives
Bring the real details
back into the picture.
When a supported AI reply echoes a replacement, Response Guard can help you restore the original for local display or copy a restored reply.
- Restore originals in your prompt, even after editing around them.
- Keep the restore mapping through page refreshes during your browser session.
- Review a reply before copying it into your next task.
Reply restore is designed for ChatGPT, Claude and Gemini. In-page restore exposes originals to the page DOM; clipboard copy is a separate action.
Restore known replacements
Your session mapping, on request
Echoed decoy
lena.fischer@example.comOriginal sample value
maya.chen@example.comRestore for local display or copy when a known replacement and session mapping are available. Reply support: ChatGPT, Claude and Gemini. In-page restore makes originals visible to the website.
Input or known replacementResult you chooseIllustrative example
03 Pro · Make it personal
Your clients.
Your projects. Your rules.
Some details matter because they matter to you. Import a CSV or JSON watchlist of exact names, emails, IDs or project codes, with a handling choice for each term.
Project OrchardRedact
maya.chen@example.comDecoy
Client referenceMask
Download the sample, edit it and import it. Watchlist contents stay on this device. Custom watchlists and regex rules are Pro features in new licensed releases.
Your watchlist
One term. Your configured action.
Protected term
Project CinderFixed decoy
Project CedarThis imported term uses a fixed decoy. Each watchlist term can instead use redact or mask. Watchlist values persist on this device.
Input or known replacementResult you chooseIllustrative example
A second look at sensitive text
From an email to an API key.
Locally bundled detectors look for familiar formats and contextual patterns while you type or paste.
People & contact
Emails, phone numbers, context-labelled names, addresses and dates of birth.
Finance & identity
Card numbers, IBANs, SSNs, passport and national ID patterns.
Secrets & credentials
API keys, tokens, passwords, credential URLs, PEM private keys and IP-address patterns.
Pro · Your own sensitive terms
Client names, project codes and IDs from your watchlist, plus your own custom pattern rules.
Detection is pattern-based and can miss details or flag something you meant to share. Review, dismiss or use your own watchlist.
It should fit the way you work
A small extension.
Room for your preferences.
A right click goes a long way.
Select text inside a protected prompt field, then redact or mask it from the context menu. Pro in new releases also offers decoys.
Auto-redact, when you choose.
Turn on optional auto-redact to scrub after a two-second pause in typing. It is off by default.
Choose your protected sites.
Use the built-in AI sites or add a compatible site yourself. Remove an origin from the allowlist to switch Shield off there.
A shortcut to a second look.
Press Alt+Shift+S for an on-demand scan. The floating shield and toolbar badge show the current finding count.
See what changed.
Review local scrub counts and recent replacements in the popup. Clear the history or remove individual records when you want.
Seven interface languages
English · Français · Español · Português · Deutsch · العربية · 简体中文
Close to home, by design
The work happens
in your browser.
Shield processes text locally and does not upload prompts, replies, watchlist values or restore mappings to Sether. New licensed clients send only account/license metadata for activation and refresh; the legacy build has no licensing requests.
Read the full privacy detailsLocal by design
Different data. Different lifetimes.
- Device
- Detection and replacements
- Session
- Restore mapping; clears on browser close
- Watchlist
- Your terms, saved until removed
The extension processes text on your device and sends no extension telemetry. Your chosen AI site receives what you submit; this picture is a concept illustration, not a network diagram.
Local feature stateSaved on this deviceIllustrative example
- In your browser
- Detection, replacement and reply review run on your device.
- For the session
- Original-to-replacement mappings can survive refreshes and clear when the browser closes.
- On this device
- Settings, watchlists, local counts and recent replacement history persist in browser storage.
Good questions
A little clarity
before you install.
Privacy is easier to trust when the details are clear.
Ask us on GitHubWhat is free, and what is Pro?
Free includes scanning, masking, token replacement and restoration of all known mappings. New official licensed releases add realistic decoys, custom watchlists and rules with Sether Pro: one person, Shield plus the editor, $5 for a month or $48 for a year, with manual renewal. Published 0.5.2 and earlier builds keep their legacy free features; check the build before starting a trial.
How does the trial work?
Install a compatible licensed build first, then explicitly start the seven-day Pro trial from your account. No card and no automatic charge. At expiry the tool returns to Free; restoring known mappings remains available, including replacements made during the trial. Session mappings still clear when the browser closes.
Does the extension send my prompts to Sether?
Prompt and reply text, watchlist contents and restore mappings stay local. New licensed clients contact Sether only when you explicitly activate or renew a license, using license metadata. No automatic polling. They do not upload those text values. The legacy 0.5.2 build does not make those licensing requests. Your chosen AI website still receives what you submit.
What stays on my device?
Your settings, site choices, custom rules, imported watchlist values, local counts and recent replacement history persist in local browser storage. Original-to-replacement mappings for restore use browser session memory. They can survive a page refresh and clear when the browser closes.
How does restoring an AI reply work?
On supported ChatGPT, Claude and Gemini interfaces, Response Guard can recognise an echoed replacement and offer local restore or copy. Show real values in this reply changes the page DOM, so the website may be able to read the restored details. Copy reply with real values writes to your clipboard instead. Restore depends on an available session mapping and the website interface.
Does Shield stop every sensitive prompt?
No detector catches everything. Review the findings and the prompt before sharing. Auto-redact is optional and off by default. In the prepared 0.5.2 update, the first send with pending sensitive findings pauses for review; sending again with the same findings proceeds. This is not a strict blocking mode.
Will it find every name or work on every website?
Names and addresses use contextual patterns, rather than a model that understands every sentence. A watchlist or a right-click action can handle particular text you want to protect. You can add sites with compatible editable fields; support can vary as website interfaces change.
Which languages are available?
The interface is available in English, French, Spanish, Portuguese, German, Arabic and Simplified Chinese. Detection includes contextual name and national ID patterns for these languages; coverage varies by format and context.
Where can I inspect the code or get help?
The Shield source and issue tracker are public on GitHub. If something looks wrong, report the browser, site and steps using fictional example data. Keep real credentials and personal details out of public bug reports.
Your next prompt, a little more considered
Good ideas deserve
a little privacy.
Free scanning, masks, tokens and known-mapping restore. See the optional Pro pilot for custom controls.