Giskard

Giskard is an open source Python library for evaluating, red teaming and generating tests for AI agents and LLM applications. Version 3 is a rewrite designed for dynamic, multi-turn testing of agents, with lighter dependencies and an asynchronous architecture.

Its scanner automatically looks for vulnerabilities in an AI application, such as hallucinations, prompt injection, harmful content or information disclosure, and turns the issues found into reusable tests. The RAG evaluation toolkit generates questions from a knowledge base to evaluate each component of a RAG pipeline. A legacy scan for tabular machine learning models remains available in version 2.

The library is open source under the Apache 2.0 licence. Giskard also offers a commercial platform for enterprises.