SafeSurf AI — Chrome Extension (Gemini Nano)

A Chrome extension powered by Gemini Nano that summarizes pages, detects phishing risks, and rewrites text locally for safer, smarter browsing.

Comment

🧠 SafeSurf AI — Chrome Extension (Gemini Nano / Gemini API)

💡 Inspiration

The idea for SafeSurf AI came from the increasing concern about how people consume online information — often without realizing the credibility or safety of the pages they visit.
As a Safety and Cybersecurity professional, I wanted to combine my background in risk assessment with my passion for technology to create a tool that helps users browse the web more safely and intelligently.

Modern browsers already offer powerful features, but with the integration of Chrome’s built-in Gemini Nano, I saw an opportunity to bring contextual AI directly to the user’s fingertips, without requiring external servers or third-party data collection.

🚀 What It Does

SafeSurf AI is a Chrome extension powered by Gemini Nano and Gemini API that enables users to:

  • Analyze pages in real time and detect potential risks or misleading content.
  • Translate text between multiple languages instantly.
  • Rewrite or proofread text to improve clarity and accuracy.
  • Operate in hybrid mode, running locally with Gemini Nano and switching to the Gemini API cloud when needed.

Mathematically speaking, the hybrid design aims to minimize latency ($L$) and maximize privacy ($P$) such that: [ \text{Optimization:} \quad \max(P) \ \text{while minimizing} \ L = f(\text{local_inference}, \text{API_fallback}) ]

🧩 How We Built It

The extension was built using:

  • Manifest V3 for Chrome Extensions
  • JavaScript ES Modules for structured logic
  • Gemini Nano / Gemini API for on-device and cloud inference
  • Chrome Storage & Messaging APIs for persistent user configuration
  • HTML/CSS minimalistic UI for speed and accessibility

The project consists of several modular layers:

  1. Background Service Worker — Handles messaging and API calls to Gemini.
  2. Popup UI — User interface for Analyze / Translate / Rewrite / Proofread actions.
  3. Content Script — Injects and extracts webpage content dynamically.
  4. Options Page — Allows enabling the “Cloud Fallback” and setting the Gemini API key.

⚙️ Challenges We Faced

The biggest challenge was integrating Gemini AI directly into Manifest V3, since the service worker environment does not support certain libraries and requires asynchronous, event-driven communication.
Additionally, we had to implement:

  • A reliable messaging system between the popup and background scripts.
  • A graceful fallback system to detect when Gemini Nano is unavailable and switch automatically to Gemini Cloud.
  • Error handling for cloud responses (404, 503) and JSON parsing stability.

These issues demanded debugging across multiple Chrome APIs and testing on ChromeOS, ensuring performance without violating sandbox restrictions.

🎓 What We Learned

Throughout the process, I learned how to:

  • Integrate Chrome AI experimental APIs (Gemini Nano) with Manifest V3.
  • Use asynchronous workers effectively for performance and privacy.
  • Design human-centered AI interfaces that focus on clarity, trust, and usability.
  • Apply software security principles to ensure no personal data leaves the user’s device unless explicitly authorized.

This project deepened my understanding of how browser-based AI can serve as a safe bridge between humans and complex information systems.

🏁 Accomplishments

  • Functional Chrome extension using Gemini Nano & Gemini API
  • Full hybrid AI workflow with local + cloud inference
  • Secure, modular codebase published on GitHub
  • Real-time risk assessment and translation capabilities

🔮 What’s Next

Future versions of SafeSurf AI will include:

  • Voice interaction via Chrome’s built-in speech API.
  • Context-aware summaries based on user intent (e.g., news vs. research).
  • Accessibility features such as text-to-speech for visually impaired users.

Ultimately, the goal is to evolve SafeSurf AI into a browser-native AI safety assistant, helping users make informed, secure, and efficient decisions online.

👤 Author

Matheus Amorim de Souza
Safety Technician & Developer | AI & Cybersecurity Enthusiast
GitHub | LinkedIn

Built With

Share this project:

Updates