Giskard Oss

Giskard is an evaluation and red teaming toolkit for large language model applications and agents. The open source side is a Python library that expresses tests as scenarios and checks: a scenario is one test case built from a message or a short conversation, and the checks around it pass or fail under pytest, so model behavior can be asserted in the same test run as the rest of a codebase. That gives engineering teams a programmatic way to catch hallucination, safety and business logic failures before a change ships. The commercial side is Giskard Hub, which takes the same testing model into a shared environment. It offers a browser interface for people who build test datasets and review results without writing code, and an SDK for developers who want to drive the same operations from Python. Hub adds shared workspaces, collaborative annotation and role based access control, centralized and versioned storage of test cases, knowledge bases that supply domain context to generated scenarios, custom failure categories and custom business checks, alerting when an issue is detected, and evaluations scheduled with cron expressions. Continuous red teaming scans deployed agents for new vulnerabilities rather than testing only at release. Enterprise controls cover single sign on, two factor authentication and audit logs. Giskard publishes no prices. The open source library is on GitHub and free to use, while Giskard Hub is sold through a conversation with the company, and no plan tiers or rates appear in the documentation or on any pricing path. The team also publishes AI safety and security research, and the open source library formed the basis of a Red Teaming LLM Applications course on DeepLearning.AI.