DEV Community

ULNIT
ULNIT

Posted on

How I Built an Autonomous Bug Bounty AI Agent on a $35 Pi

The $5 Goal

My goal was simple: earn money legally through bug bounty, using nothing but a Raspberry Pi 4B (1GB RAM) and an AI agent — no cloud, no expensive tools, no manual work.

The Stack

  • Hardware: Raspberry Pi 4B (arm64, Debian)
  • AI Agent: Hermes Agent (open-source, self-improving)
  • Tools: Pure Python stdlib + curl + nuclei
  • Targets: HackerOne & Bugcrowd programs

What Worked

GraphQL Schema Enumeration

Found a misconfigured Apollo Server where field suggestions bypassed introspection protection — enumerated entire schema through error messages.

Internal Service Discovery

CSP headers are treasure maps. Parsed Content-Security-Policy on public pages to discover internal microservices, staging CDNs, and backend APIs.

Subdomain Reconnaissance

Built a zero-dependency pipeline using crt.sh, AlienVault OTX, URLScan — discovers subdomains, filters catch-all servers, probes live services.

What Didn't Work

  • Config.js token scanning (post-Interseller era)
  • CORS misconfigurations (informational, no bounty)
  • Stack trace exposure (P5 at best)

Key Lessons

  1. Depth beats breadth — one deep target > 10 shallow scans
  2. Legacy infra is gold — web hosts still run Apache 2.2.3
  3. Write reports immediately — cookies expire
  4. P4-P5 doesn't pay — go for XSS, SSRF, IDOR

Open Source Tools

Built with ❤️ on a Raspberry Pi. Get the tools →

Top comments (0)