Icebreaker is an open-source AI agent fine-tuned for active cybersecurity operations — analyzing targets, reasoning through attack vectors, and executing multi-step exploits without human prompting.

Model Details


Base Model	Qwen 2.5 7B
Training Platform	Kaggle
Fine-Tuning	Supervised Fine-Tuning (SFT)
Datasets	ToolBench · CTF Write-ups
Format	GGUF (local inference)

Fine-Tuning Datasets

ToolBench — Trains the agent on advanced tool manipulation and API calling, enabling seamless interfacing with penetration testing utilities.
CTF Write-ups — Trains offensive security reasoning, vulnerability analysis, and multi-step exploit chaining.
Primus-Reasoning - Cybersecurity reasoning
Bug-Bounty-pentest-en - Methodologies (OWASP, PTES), checklists by app type, attack techniques, platforms, report templates and tools.

Core Capabilities

Autonomous Exploitation — Analyzes environments, identifies attack vectors, and executes exploits independently.
Dynamic Tool Use — Interfaces with tools like Nmap and Metasploit, interprets CLI output, and adapts strategy in real time.
Air-Gapped Operation — Runs entirely on-premise via GGUF. No cloud API calls, no data leakage.
Open-Source & Extensible — Designed for community-driven dataset contributions and custom tooling integrations.

Intended Use

Icebreaker is intended for:

Authorized penetration testing engagements
CTF competitions
Security research and red team simulation
Offline/air-gapped environments handling sensitive infrastructure data

This model is for authorized security testing only. Misuse against systems without explicit permission is illegal and unethical. The authors are not responsible for misuse.

License

Apache 2.0

pannagkv/icebreaker-v2

Icebreaker

Model Details

Fine-Tuning Datasets

Core Capabilities

Intended Use

License

pannagkv/
icebreaker-v2