🛡️ CyberSentinel — Autonomous Network Threat Detection Agent

Inspiration

Cybersecurity threats are growing every day, yet most small organizations and individuals have no affordable way to monitor their own networks. Enterprise-grade tools cost thousands of dollars monthly, and manual monitoring is simply not realistic — attackers don't sleep, but humans do.

I was inspired to build something that could watch, think, and respond autonomously — a digital security guard that never takes a break.

What I Learned

• How autonomous AI agents work in a real-world context
• How to integrate GitHub Models API (GPT-4o-mini) into a Node.js backend
• How TCP port scanning works at a low level using Node's net module
• How to build a real-time dashboard that auto-refreshes threat data
• The importance of ethical hacking boundaries — only scanning systems you own or have permission to test

How I Built It

CyberSentinel is built around an autonomous agent loop:

$$\text{Scan} \rightarrow \text{Analyze} \rightarrow \text{Classify} \rightarrow \text{Log} \rightarrow \text{Repeat}$$

1. Scanner — A pure Node.js TCP port scanner checks 12 critical ports every 60 seconds on the target host
2. Analyzer — Scan results are sent to GPT-4o-mini via the GitHub Models API, which reasons about open ports and returns a structured threat assessment
3. Responder — Threats are classified as LOW, MEDIUM, or HIGH and stored in a JSON log with timestamps
4. Dashboard — A live HTML/CSS/JS dashboard displays all threats in real-time, auto-refreshing every 10 seconds

The entire system runs with a single command:

node server.js

Challenges I Faced

• nmap not available on Windows — I replaced it with a pure Node.js TCP socket scanner that works cross-platform
• GitHub Models API authentication — Fine-grained tokens don't support the Models API; classic PATs are required
• Getting structured AI responses — Prompting GPT-4o-mini to return consistent JSON required careful system prompt engineering
• Building fast under time pressure — The entire project was built in under 30 minutes during a live hackathon session

Built With

• css
• dotenv
• express.js
• github-models-api
• gpt-4o-mini
• html
• javascript
• node.js