Welcome to LLMBUS!
Your retro-cyberpunk companion for developers, security pros, and curious AI explorers.
Step into a world where neon lights flicker, synth beats pulse, and it’s always 1999 somewhere...
| Category | Capability | Description |
|---|---|---|
| Transformation Utilities | ROT13 · Morse Code · Pig Latin · ASCII-Art | Instantly convert, obfuscate, or stylise your prompts. |
| Render Text | Styled Preview · Image & Audio Export | See your text rendered, then download it as an image or audio. |
| Background & Text Color | Customize colors before exporting as image. | |
| Download Audio | Convert text to speech (TTS) and save it. | |
| Tokenizer | Tokenizer Viewer | View token boundaries for popular LLMs. |
| Workflow | Project Tracker | Organize experiments, manage prompts, and track outputs. |
| Portability | Import / Export (JSON) | Save and version-control entire project files. |
| Search | In-App Search | Quickly find any prompt. |
| Paraphrase | Text Rewriter | Rewrite text with new words, same meaning. Uses phi-4:latest by default. |
| Fuzzing | Bus Driver (Neo-GPT / Ollama Fuzzer) | Sends crafted prompts to LLMs using Nano-GPT or Ollama for fuzzing and robustness testing, with up to three attempts per prompt. If the model fails to block a malicious prompt even once, the entire prompt is marked as failed. The default Ollama model is phi4:latest, but you can replace it with any model name you prefer. If Nano-GPT settings are not configured, Ollama will be used instead. At the end of the process, it automatically generates a detailed report of the results. |
-
Install the LLMBUS Chrome Extension (required).
-
If you’re running Ollama in Docker and encounter CORS errors, use the following command:
sudo docker run -d --gpus all \
-v ollama:/root/.ollama \
-p 11434:11434 \
-e OLLAMA_HOST=0.0.0.0 \
-e OLLAMA_ORIGINS="*" \
--name ollama1337 ollama/ollamaBig thanks to these amazing resources that inspired parts of LLMBUS:
-
An evolving reference on testing AI systems for security and robustness.
-
Garak (NVIDIA)
A fuzzing tool for LLMs, useful for red-teaming and model evaluation. -
Deck of Many Prompts
A creative prompt generation tool for AI writing and experimentation. -
Parseltongue
A powerful prompt hacking tool/browser extension. -
ArXiv: 2411.01084
Plentiful Jailbreaks with String Compositions -
ArXiv: 2412.03556
BEST-OF-N JAILBREAKING
This tool is strictly intended for ethical and research purposes only.
Any form of misuse, including malicious activities, is strictly prohibited and may result in legal consequences.
Users are expected to act responsibly and in accordance with ethical and legal standards when using LLMBUS.