LLMVault
A deliberately vulnerable, CTF-style training platform for practicing attacks and defenses from the OWASP Top 10 for LLM Applications, in a scored range or against a real local model.
LLMVault is a deliberately vulnerable, CTF-style training range for the OWASP Top 10 for LLM Applications, intended for authorized, self-hosted security education.
Play Mode is a scored range of 25 labs across three tiers: ten core labs covering one category each, ten advanced multi-turn labs covering jailbreaking, data poisoning, agent exploitation and model extraction, and five expert labs modeling real-world attack classes, with each tier unlocking the next and scripted assistants so flags reproduce reliably. Live Mode points the same attacks at a real local model run through Ollama, with a secret generated fresh each session and no fixed flag to look up. Every lab pairs an attack with the defense that would have stopped it.
It includes a dashboard tracking progress, rank and score across both modes, and is meant to run only in a controlled environment, not exposed to the internet.