Sentinel

An autonomous browser automation agent with real-time monitoring and error recovery capabilities.

Overview

Sentinel is an agentic system that uses Gemini 2.5 Pro with Thinking Mode to autonomously navigate web pages and recover from errors. Unlike simple retry logic, Sentinel reasons through navigation failures—diagnosing issues and adapting its approach in real-time.

This project was built with AI safety principles in mind: agentic systems need observability baked in, not bolted on. The monitoring dashboard provides real-time visibility into agent decision-making and actions.

Features

• Autonomous Error Recovery: Agent reasons through navigation failures and adapts its strategy
• Real-Time Monitoring Dashboard: React-based UI for observing agent actions as they happen
• Thinking Mode Integration: Leverages Gemini 2.5 Pro's reasoning capabilities for robust decision-making
• Browser Automation: Full browser control via Puppeteer

Tech Stack

• Frontend: React, TypeScript
• Backend: Node.js, Puppeteer
• LLM: Gemini 2.5 Pro with Thinking Mode

Getting Started

Prerequisites

• Node.js
• Gemini API key

Installation

1. Clone the repository
2. Install dependencies: npm install
3. Create a .env.local file and add your Gemini API key
4. Run the application: npm run dev

Project Structure

• backend/ — Node.js server + Puppeteer browser control
• components/ — React UI components
• services/ — LLM integration + agent logic
• App.tsx — Main application
• index.tsx — Entry point

Why This Matters

As AI agents become more capable, oversight and observability become critical. Sentinel demonstrates that monitoring can be a first-class feature of agentic systems—not an afterthought.

Author

Lesley Ward — GitHub