AI Prompt - Security Guard — Stop accidentally sending secrets to AI chatbots

// The whole point

Private by design.

A tool that protects your secrets shouldn't become the thing collecting them. So it doesn't — at all.

Scans locally No prompt logging No telemetry No accounts No network calls

How it works

Three steps, zero friction.

Type or paste as usual

Use ChatGPT, Claude, Gemini, Perplexity or Copilot exactly how you already do. Nothing to learn.

It catches risks locally

A live badge watches the box. The moment something sensitive appears, it flags it — on your device.

Redact, edit, or send

One click swaps secrets for safe labels like [API_KEY]. You're always in control.

The magic moment, in motion

demo reel: type → badge → warning → redact → send safely

Why it matters

It's not a feature list. It's a save.

Before using AI at work

Catch client names, internal plans, financials and legal or health info before they reach a chatbot.

client datafinancialslegal / health

Before pasting code

Catch API keys, tokens, credentials and secrets hiding in stack traces, configs and snippets.

api keystokenscredentials

Before uploading documents

Scan PDFs and Word files locally — including the hidden comments and metadata you forgot were there.

pdf / docxcommentsmetadata

Features

Calm protection that stays out of the way.

Local scanning

Analyzed on your device, never stored or uploaded.

Pre-send warning

Lists exactly what was found — masked values only.

Live risk badge

A small badge near the input updates as you type.

One-click redaction

Swaps secrets for labels like [EMAIL] and [API_KEY].

Attachment scanning

Reads PDF & Word text locally, comments and metadata too.

Three sensitivity modes

Basic, Balanced and Strict, switchable any time.

Shield Mode

Optional private composer: sites can't read your draft as you type — text stays yours until you approve it.

Per-site control

Turn watching on or off per site; add custom domains.

Accessible

Keyboard operable, screen-reader ready. WCAG 2.1 AA & Section 508.

Least privilege

Runs only on supported AI sites; requests only what it uses.

How it compares

Built to catch what other approaches miss.

Capability	AI Prompt - Security Guard	Manual checking	Cloud-based tools
Scans before you send	Yes	No	Sometimes
Runs entirely locally	Yes	Yes	Usually no
No account required	Yes	Yes	Often no
One-click redaction	Yes	No	Varies
No prompt logging	Yes	Yes	Varies

Trust & verification

Don't take our word for it. Verify it.

For a privacy tool, credibility is the product. Everything you need to check the claims yourself.

View source on GitHub Chrome Web Store listing

How we verify "no network calls"

Open DevTools → Network while you type and send. You'll see the extension make zero requests.

Version & changelog

v1.0.0 — initial release. Release notes ship with every store update.

Requested Chrome permissions

storage: Save your settings & the counter — on your device.
offscreen: Run the bundled PDF text extractor locally. No requests.
scripting: Run the bundled scanner on custom domains you add — never remote code.
host (AI sites): Only the AI tools you enable — never all sites. Custom domains are granted one origin at a time, when you add them.

No access to browsing history, bookmarks, or downloads.

Who should install this?

Developers pasting stack traces, logs, configs or code.
Consultants working with client documents.
Employees using AI with internal company data.
Students & researchers handling participant or unpublished work.

…and anyone who has ever pasted first and panicked afterward.

Free bonus

The one-page guide: what you should never type into an AI chatbot.

Five categories to keep to yourself, smart habits, and one golden rule. A handy poster to pin up — the extension does the catching for you.

Download the PDF Open in a new tab

Poster: What you should never type into an AI chatbot — five categories to keep to yourself, smart habits, and the golden rule.

Stop accidentally sending secrets to AI chatbots.

Before you send this