GEMMA-by-GOOGLE

GEMMA-POWERED-BY-GOOGLE-CYBERSECURITY-AUTONOMOUS-AI:

An Autonomous AI agent inside of Linux environment with one of the worlds most cutting edge AI models, Googles GEMMA 4-12b Model. Fully uncensored/Abliterated. FULLY

LOCAL. FULLY FREE. With PERSISTENT negative cache learning, adaptation.Learning and self harnessing getting more self aware and intelligent with each engagement.

Autonomous recon, scanning and attack vector mapping, - one word to start it all: ENGAGE. Attack-loops, reports, professional and comepletely local -this agent is fast, and
documents its exploits, findings and risk levels autonomously with clean precision and professionalism. Star if you like it, or open a PR and lets do something together. This is where ideas come alive and problems are solved!!

license: mit
language:

• en
  tags:
• security
• penetration-testing
• autonomous-agent
• mcp
• kali-linux
• llm
• cybersecurity
• red-team
  library_name: other
  pipeline_tag: text-generation

🔐 HALO Cybersecurity

Autonomous AI-powered penetration testing agent — fully local, no cloud, no API keys.

Built on Kali Linux with a local LLM (Gemma Powered by Google via LM Studio) and a Flask-based MCP tool server. The agent runs recon, attacks, and generates professional pentest reports — all autonomously.

What It Does

• 🔍 Autonomous recon — masscan + nmap to discover open ports and services
• ⚔️ Autonomous attack loop — selects and chains tools based on what it finds
• 🧠 Persistent negative experience cache — learns what fails across ALL sessions and never wastes time on it again
• 📝 Auto-generates branded HTML pentest reports on session end (Ctrl+C)
• 🔒 100% local — Gemma4-12B running in LM Studio, nothing leaves your machine
• 🧩 Adaptive skill injection — loads relevant attack playbooks into the prompt based on the current goal
• Self aware and self correcting harnessing

Tool Arsenal (23 Tools)

Architecture

agent_loop.py  ──►  mcp_server.py (Flask, port 8000)  ──►  security tools
     │
     ├──►  agent_cache.py       (persistent negative experience cache)
     └──►  report_generator.py  (auto HTML pentest report on exit)

Sovereign Agent Layer v1

The negative experience cache fingerprints every tool call. If it fails once, it gets one retry. Fail twice — permanently blacklisted and the agent subsequently moves on to next, more practical tool for the job. The agent never wastes cycles on dead ends it has already proven don’t work. Instead, the agent autonomously structures its learning through trial and error harnessing where it learns what will and will not work for each particular attack. If success, the agent prints a thumbs up to the user, denoting said success. Then arrives at the next attack mission.

Stack

• Model: Gemma4-12B Instruct Abliterated (GGUF via LM Studio)
• Agent: Python autonomous loop with MCP tool calls
• MCP Server: Flask on port 8000
• OS: Kali Linux (UTM on Apple Silicon M1)
• Hardware: MacBook Pro M1 16GB RAM

Usage

cd /home/bigkali/security-agent
python3 agent_loop.py

>>> engage 192.168.64.3    # full autonomous recon + attack
>>> run nmap on 10.0.0.1   # single goal query
>>> exit                   # triggers HTML report generation

Active development. New capabilities and upgrades pushed regularly.

Built by a self-taught developer and security researcher. One year in.

• DISCLAIMER *
• This is a community project designed by independent developer and is not affiliated with or sponsored by the Google corp.

⚠️ Content Warning: This model is heavily abliterated and will respond
to sensitive or explicit requests without restriction. Not suitable for
minors or unmonitored environments. Use responsibly and legally.

🔒 Legal Warning: This tool is intended strictly for authorized
penetration testing and security research on systems you own or have
explicit written permission to test. Unauthorized use is illegal.

License

MIT