FireGuard

Browser Extension

Protect ChatGPT, Claude, Gemini, Copilot, Perplexity, and DeepSeek prompts in the browser with FireGuard input guardrails.

Use the Fireraven browser extension when you want FireGuard to check prompts on consumer AI chat sites before they reach the model. The extension runs input guardrails only.

Supported Sites

ChatGPTchatgpt.comGoogle Geminigemini.google.comClaudeclaude.aiPerplexityperplexity.aiMicrosoft Copilotcopilot.microsoft.comDeepSeekchat.deepseek.com

Supported Browsers

ChromeEdgeBraveFirefox (planned)Safari (planned)

Chrome, Edge, and Brave install from the same Chromium store listing. Firefox and Safari listings come later. Activation emails include a download button for each browser that has a store URL configured.

Get an Activation Code

Platform users

  1. Open the Fireraven app.
  2. Go to Connectors → Browser extension.
  3. Click Generate new activation code and enter the email that should receive the code.

End users

Organization owners invite end users from Settings → Organization → End Users. Invitees receive the same activation email (one-time code; no platform login required).

Install and Activate

  1. Install the Fireraven browser extension for Chrome, Edge, or Brave.
  2. Open the extension → Set up credentials → Activate with code.
  3. Enter the email and one-time activation code from the email. Codes expire (default 7 days) and work once.
  4. Credentials are stored in the browser. FireGuard shows as connected.

Paste Credentials

If you already have platform access, you can skip the activation code:

  1. Open the extension → Set up credentials → Paste credentials.
  2. Enter your FireGuard API key and project ID (optional organization ID).

See FireGuard API keys and Project configuration.

Organization Settings

Configure org-wide extension behavior in the Fireraven app under Connectors → Browser extension. Settings are scoped to the organization; until you save, defaults apply.

SettingAPI fieldDefaultDescription
Allow end users to override a blockallowEndUserOverridefalseWhen enabled, end users can dismiss a FireGuard block in the extension and send the prompt anyway. When disabled, blocked prompts cannot be overridden.
PII redact for files and images — Text content onlyfilePiiRedactMode: text_onlyyesRedact PII in real text content only. Leave image and embedded non-text as-is, and send redacted text plus original files and images to the LLM.
PII redact for files and images — Extract and redactfilePiiRedactMode: full_extractnoExtract text from files and images, redact PII, and send only the redacted extracted text to the LLM.

The extension can read these settings with a FireGuard organization API key. The shipped extension does not enforce them yet; see Get browser extension settings.

curl \
  "https://api.fireraven.ai/public/connectors/browser-extension/settings" \
  -H "X-API-Key: $FIREGUARD_API_KEY"

Example response (defaults when no org row exists):

{
  "allowEndUserOverride": false,
  "filePiiRedactMode": "text_only"
}

What It Does

On Send / Enter, the extension pauses the outbound chat request, calls FireGuard input_guardrails with the current-turn text and any resolved images / documents, then:

ResultBehavior
UnsafeBlock the send
SafeAllow the original prompt
Out of creditsAllow and show a warning toast
Not connected, timeout, or API/bridge errorAllow and show a toast (connect or contact support)
Unsupported / unresolvable attachmentsAllow text (and any resolved media) through; toast that some attachments were skipped

Media is taken from the current user turn (upload cache + chat payload), not from the full chat history. When images or documents are present, the extension sends multipart input_guardrails (payload JSON plus binary images / files parts) — not base64 in JSON. FireGuard accepts up to six image/* parts and documents PDF (.pdf), Word (.docx), Excel (.xlsx), plain text (.txt), and Markdown (.md) (see Usage & billing).

Output guardrails are not applied by the extension today.

Notes

  • Input guardrails only — model output is not checked by the extension.
  • Fail-open on infrastructure or credential problems; only policy/security unsafe results block.
  • Users can disable or remove the extension in the browser.
  • If a site changes its request format and the prompt cannot be extracted, the send may proceed without a FireGuard check.

On this page