Free AI Safety Workshop — in collaboration with explainx.aiSave your seat
Agentbeam
Free Chrome extension

Catch secrets before they reach the model.

The Beam Sentinel Chrome extension checks what you’re about to send to ChatGPT, Claude, Gemini, and other AI chat assistants — before it leaves your browser. Free, no account, nothing sent anywhere.

  • Free, full stop
  • No account or sign-in
  • No network request
  • Local checks only
Beam Sentinel introduction: checking prompts before they're sent to an AI chat assistant

What it catches

One pause, right before the click that matters

Pasting a config, a stack trace, or a customer record into an AI chat takes one click. Sentinel adds a half-second pause exactly when that click looks risky.

Beam Sentinel dialog blocking a message that contains private information

Private data, caught before send

Email addresses and other labeled personal data pause the send, with a chance to review before anything goes out.

Beam Sentinel toast confirming a secret was auto-redacted from a message

Live secrets, removed automatically

A live-looking API key, access token, or private key block gets stripped out of the message in place, and the send is held so you can check the cleaned version first.

Beam Sentinel coverage grid listing the six risk categories it checks

Six categories of risk

Personal data, private keys, outbound credentials, destructive commands, instruction overrides, and credential references — checked on every send.

Beam Sentinel warning toast alongside its local-first privacy points

Runs entirely on your device

No account, no sign-in, no network request. Checks and redaction happen inside the extension — Sentinel never sends your prompts anywhere, including to Beam.

How it works

Four steps, all inside the browser

  1. Runs where you already chat

    Works automatically on the major AI chat assistants — nothing to configure after install.

  2. Checks right before it sends

    The message is read the instant you hit Send, Submit, Run, or press Enter — never before.

  3. Acts on what it finds

    A live secret gets redacted and the send held. Other high-risk findings pause for “Keep blocked” or “Allow once.” Lower-risk findings get a short warning and continue.

  4. Stays out of the way

    Ordinary prompts go through with zero friction — Sentinel only interrupts when something actually looks risky.

Beam Sentinel toolbar popup showing protection status and recent actions

Free and open about what it does

Nothing to sign up for, nothing hidden

Sentinel only requests the storage permission — used to remember your on/off setting and a local log of the last 100 redacted decisions. No browsing history, no other tabs, no network permission. Read the full breakdown in the privacy policy.

Price
Free, no paid tier
Account required
None
Network requests
None
Permissions requested
storage, only
Data retained
Last 100 redacted decisions, locally
Raw prompt content stored
Never

Add it in under a minute

Free, from the team behind Agentbeam. No account, nothing to configure.

Add to Chrome — free