H helmian guard

Browser extension for AI oversight

Give risky AI actions a second look.

Helmian Guard helps you catch potentially harmful, evasive, or hedging AI behavior in the browser—before an answer becomes an action.

View package statusSee how Guard works

v0.2.3 · Chromium and Firefox package paths are staged for review; no store submission yet.

Packages & access

Builds stay reviewable before they become downloads.

Each browser target has its own manifest and signing boundary. The page never exposes an EXE or invents an entitlement while artifacts or billing are unconfigured.

Buy Guard · $10 one-timeCheckout opens Stripe securely.

CHROMIUM MV3

Chrome · Edge · Brave · Opera · Vivaldi

One reviewed Chromium package, loaded unpacked for testing before store distribution.

Artifact pending review
Download package

FIREFOX WEBEXTENSION

Firefox

Separate manifest and AMO signing path. The Gecko add-on ID is not yet owned.

AMO signing required
Download package

Install

No Chrome Web Store listing yet—so it's a manual load.

Same as installing any extension in developer mode. No code runs until you do this.

  1. 01

    Download & unzip

    Get the zip above, then extract it anywhere on your computer.

  2. 02

    Open chrome://extensions

    Turn on Developer mode (top right), then click Load unpacked.

  3. 03

    Select the unzipped folder

    Guard is active immediately on claude.ai, ChatGPT, Gemini, and Grok. No account, no network calls.

What Guard notices

Not every concern is a violation.
Every concern deserves context.

Guard is designed to surface signals for human review, not to pretend it can read intent with certainty.

01

Harmful moves

Flags content that appears to steer toward harm, abuse, or unauthorized outcomes.

Pause before proceeding

02

Hedging & evasion

Highlights language that sidesteps accountability, obscures consequences, or buries an actionable step.

Ask for clarity

03

Scope drift

Draws attention when an AI response moves beyond the request, authority, or stated boundary.

Confirm the boundary

How it works

A small pause. A clearer decision.

Guard stays close to the conversation and makes its concern legible when it finds one.

  1. 01

    Read the exchange

    Guard evaluates the visible AI interaction for patterns that warrant attention.

  2. 02

    Show the signal

    It identifies the concern in plain language, without overstating confidence.

  3. 03

    Keep you in control

    You decide whether to revise, stop, investigate, or continue with context.

Our principles

Signal, don’t sensationalize.

Useful oversight should be candid about what it knows and careful about what it does not.

Helmian Guard

Keep the human
at the point of action.

Built for people who want more context around AI behavior, right where it happens.

Back to top