1. What FirstGuard does

FirstGuard is a browser extension that locally analyzes content you type into large language model (LLM) interfaces to detect potential sensitive data leaks (personal data, secrets, financial, medical, or legal information) before submission.

The analysis relies on four complementary layers, all executed locally:

• Regular expressions — Detection of known patterns (social security numbers, IBANs, API keys, etc.).
• Aho-Corasick dictionary — Linear-time search across ~5,800 sensitive terms (drugs, conditions, legal and financial terms, FR + EN).
• NER Piiranha (opt-in) — AI-powered named entity recognition model, running in an offscreen document. The ONNX model (~280 MB) is downloaded once from a public CDN (jsDelivr) then cached locally in IndexedDB.
• OCR Tesseract.js — Text extraction from attached images (PNG, JPG, WEBP, BMP, TIFF) directly in the browser, with no server upload.

2. Data collected

No personal data is collected, transmitted, or stored on any external server.

• All analysis is performed locally in your browser.
• The raw text of your prompts never leaves your device.
• Images scanned by OCR are processed in memory and are neither stored nor transmitted.
• Scan history (containing only partially redacted excerpts) is stored in chrome.storage.local, on your device only.
• Your preferences (alert thresholds, active rules) are synced via chrome.storage.sync across your Chrome-connected devices.

3. Network connections

FirstGuard contacts no proprietary server. The only network connection occurs when NER detection is enabled (opt-in): the ONNX model is downloaded from cdn.jsdelivr.net on first use, then cached in IndexedDB. No user data is sent during this download.

4. Permissions requested

• storage — Local scan history and preference synchronization.
• tabs — Identify the active tab to deliver scan results.
• activeTab — Access active tab content for prompt interception.