Giskard is a French company offering testing and red-teaming for AI agents and language models. Its open-source library and the commercial Giskard Hub test conversational agents as a black box, needing only an API endpoint. They look for security issues such as prompt injection and data disclosure, and for quality issues such as hallucinations and inappropriate refusals. Test cases can be generated from sources such as RAG knowledge bases, and domain experts can contribute to annotation and red-teaming through a collaborative environment. Giskard also offers managed services, including a go/no-go report.
Giskard is a European company with France as its country of establishment. For managed hosting of the Hub, the website says you choose between EU and US data residency; choose the EU if data must stay in Europe. For sensitive sectors such as the public sector and defence, the vendor says an on-premises installation is available. You can run the open-source library yourself. The vendor states a policy of not using customer data for training.