Prompt Fuzzer

What Prompt Fuzzer is

Prompt Fuzzer is a free, open-source tool for evaluating and improving the system prompts used by AI applications. It generates varied scenarios and uses malicious prompt variations as targets for dynamic LLM-based attack testing. Developers can run campaigns in interactive chat mode, from the command line, or with multithreaded testing. Its Playground supports successive chat iterations to adjust settings, with each iteration forming a new fuzzing campaign. The repository lists 16 attack types, including jailbreak, prompt injection, RAG poisoning, and system prompt extraction, and support for 16 LLM providers, including OpenAI, Anthropic, Azure OpenAI, Cohere, Google PaLM, and Ollama. Custom and self-hosted endpoints are supported, including Ollama and OpenAI-compatible endpoints. The tool is obtained through GitHub and installed with pip; it requires Python 3.10 or newer. RAG testing needs an embedding provider and model, and ChromaDB is installed by default with the package. Prompt Fuzzer is self-hosted, and the maker warns that use consumes tokens. Its focus is prompt-based systems and how they interact with a target program, rather than symbolic execution of code paths.