KawaiiGPT is an open-source framework for analyzing jailbreak behavior in large language models. The project provides a unified interface for interacting with multiple backend LLMs through a custom reverse-engineered API wrapper. It is designed for security research, red-team testing, and LLM safety evaluation.

• Unified LLM Gateway Connects to multiple backend models (DeepSeek, Gemini, Kimi-K2) through a custom reverse-engineered API wrapper.

• Jailbreak Evaluation Suite (WormGPT-Class Capabilities) A controlled testing environment replicating key WormGPT-style behaviors for research:

  • analysis of unrestricted LLM outputs under jailbreak pressure,
  • behavioral deviation mapping under safety bypass attempts,
  • generation of unfiltered model completions for alignment benchmarking,
  • payload-structure analysis (scripts, templates, automation patterns) to study LLM misuse scenarios,
  • comparative testing across multiple models for identifying guardrail weaknesses.

• Prompt Injection & Safety Boundary Testing Tools for studying:

  • prompt-extraction attempts,
  • refusal-bypass strategies,
  • role, system, and persona hijacking techniques.

• Context-Free and API-Key-Free Execution All operations run locally without external credentials.

• Automated Script Sandbox Evaluates LLM-generated code (Python/JS/Bash) in a safe, isolated environment for research on:

  • code-generation reliability,
  • script hallucinations,
  • error propagation,
  • model-driven automation patterns.

• Communication Template Generator (Research Mode) Used to study automated message creation patterns:

  • email structures,
  • organizational tone mimicry,
  • text-style reproduction,
  • linguistic fingerprinting of LLMs.

• Model Behavior Profiling Generates structured reports identifying:

  • risk scores,
  • guardrail bypass likelihood,
  • output volatility,
  • jailbreak susceptibility.

You can run KawaiiGPT on Windows using the prebuilt executable provided in the Releases section.

1. Open the Releases tab of this repository.
2. Download the latest file: KawaiiGPT.exe

On first run the application will:

• create a local configuration folder,
• generate default settings,
• initialize the LLM routing layer.

No installation of Python or Git is required.

KawaiiGPT uses a reverse-engineered request layer that:

• routes prompts to backend LLM servers,
• provides uniform input/output formatting,
• supports predefined jailbreak patterns for research purposes.

A built-in menu offers:

• prompt-injection examples,
• output comparison across models,
• analysis of guardrail bypass attempts.

• Red-team assessments
• Security audits of AI-driven systems
• Research on model alignment, prompt-injection, and safety failures
• Educational demonstrations of dual-use risks

This project must only be used for authorized security testing and academic research.

• This project was made to be fun!
• All risks or consequences that you have done are your own responsibility.
• Changing or selling this tools is prohibited and not allowed!. ⚠
• Using a prepared model, not a fine-tuned model!
• All LLM here uses a prompt injection! (jailbreak has been written on help-menu)
• This is KawaiiGPT, not WormGPT, made to be fun or companies!
• WormGPT tag only used for jailbreaked model

"Why is it obfuscated?"

Me: Okay, many people asking this question, the reason why i obfuscate the
    code is because I want to avoid recoding and renaming which ends up 
    selling KawaiiGPT tools under my name and claiming that it belongs to them

"This is obviously a virus or a RAT"

Me: Seriously, i'm enough with this question over time. No, there is no RAT, Spyware,
    Malware, Ransom, etc. Why would i do that? because it's free and obfuscated
    doesn't mean i wanna trap you all, **i wouldn't**. If that wasn't from me
    don't run it or you would fall in a trap. 

"But why is it obfuscated?"

Me: Like what i said earlier, "I want to avoid recoding and renaming which ends up 
    selling KawaiiGPT tools under my name and claiming that it belongs to them".
    Please understand that i wouldn't put a malicious code into my tools/code,
    if somebody said "KawaiiGPT has a virus!!" in Telegram or other please don't
    trust it, maybe they just want to bring down my name or they got the wrong KawaiiGPT
    that is not from me.