A research workbench for developing and testing attacks against large language models, with a focus on prompt injection vulnerabilities and defenses. - View it on GitHub