Giskard

Giskard is an open-source testing framework for AI models that automatically surfaces vulnerabilities, biases, and quality problems in LLMs and traditional ML models. It supports evaluation, red-teaming, and test generation for agentic systems with a modular, async-first design, and includes RAG evaluation and LLM security testing. Optional aggregated telemetry can be disabled. Released under the Apache 2.0 license.