Kali-Terminus-v1 Model Card

Kali-Terminus-v1 is a specialized, lightweight function-calling language model (~270M parameters) fine-tuned specifically for Linux terminal execution, cybersecurity tool invocation, and automated penetration testing workflows in Kali Linux.

It is created by fine-tuning google/functiongemma-270m-it on the iselabvn/Kali-terminal-executor dataset of terminal tool calls using LoRA/QLoRA, and subsequently merging the adapter weights back into the base model to deliver a fast, standalone inference experience without requiring external PEFT dependencies.

---

Model Summary

• Developed by: ISeLab
• Model Type: Causal Language Model (Gemma-based Function Calling)
• Base Model: google/functiongemma-270m-it
• Dataset: iselabvn/Kali-terminal-executor
• Parameter Count: ~270 Million
• Precision: Float32 / BFloat16
• Context Window: 2048 tokens
• License: Gemma Terms of Use

---

Intended Use

Primary Use Cases

1. Automated Security Agent Execution: Serving as an execution worker/tool-calling agent inside a multi-agent architecture (translating task instructions into precise, structured command execution payloads).
2. Kali Terminal Assistant (Single-Turn / Multi-Turn): Invoking CLI tools (nmap, gobuster, metasploit, hydra, sqlmap, etc.) with correct flags and parameters based on tool definitions.
3. Edge Deployment: Due to its compact 270M parameter size (~1GB disk footprint), kali-terminus-v1 runs efficiently on low-resource environments, local VMs, or edge security appliances.

[!IMPORTANT] Usage Guidelines & Recommendations:

• Recommended Roles: Designed specifically to act as an Agent Executor within a Multi-Agent System (where a primary planner agent orchestrates goals and delegates execution tasks to this model) or as a specialized Single-Turn Kali Terminal Assistant.
• Not Recommended: Standalone usage as an open-ended general chat assistant or unguided autonomous strategist.

Out-of-Scope & Misuse

• Standalone deployment as a general-purpose chat assistant without structured tool-calling context.
• Unauthorized computer access or offensive operations against targets without prior explicit written permission.

---

Quickstart & Usage

Standard Transformers Inference

You can load and run kali-terminus-v1 directly using Hugging Face transformers:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "kali-terminus-v1" # or local path "./kali-terminus-v1"

# Load tokenizer and model
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.bfloat16 if torch.cuda.is_available() and torch.cuda.is_bf16_supported() else torch.float32

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=dtype,
    device_map=device,
    trust_remote_code=True
)

# Example prompt
prompt = "Perform an Nmap stealth scan on target 192.168.1.100 for top 100 ports."

inputs = tokenizer(prompt, return_tensors="pt").to(device)
with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=128, do_sample=False)

response = tokenizer.decode(outputs[0], skip_special_tokens=False)
print(response)

---

Training Details

Training Dataset

• Dataset Repo: iselabvn/Kali-terminal-executor
• Description: Multi-turn and single-turn Linux terminal execution datasets covering tool declaration, argument validation, command execution, and stdout/stderr observation loops for cybersecurity tools and administrative tasks in Kali Linux.

---

Safety & Ethics Disclaimer

kali-terminus-v1 is designed exclusively for authorized penetration testing, security auditing, educational research, and administrative automation. Users are solely responsible for ensuring compliance with all applicable local, national, and international laws when executing commands generated by this model.