Snowglobe

Snowglobe is Guardrails AI’s simulation product for teams building conversational AI agents. Developers connect an agent through an API or SDK, configure personas and scenarios, and generate conversations to uncover failure patterns before deployment. The resulting judge-labeled datasets support evaluation and fine-tuning, including preference pairs and supervised training examples. Teams can also save test suites for regression testing and track error rates across builds.

Guardrails AI was founded in 2023 by Shreya Rajpal, Diego Oppenheimer, Safeer Mohiuddin and Zayd Simjee, with Rajpal serving as CEO. Introduced in 2025, Snowglobe grounds simulated personas in an agent’s context and orchestrates stateful, multi-turn conversations. Its scenarios include ordinary user behavior rather than assuming every interaction is adversarial. This complements Guardrails AI’s work on reusable validators that check application inputs and outputs against defined policies.

In 2025, the company reported that customers were generating tens of thousands of simulated conversations before production. MasterClass used Snowglobe for synthetic data, while Changi Airport Group used it to test hallucination and toxicity risks. Guardrails AI’s business milestones include a $7.5 million seed financing led by Zetta Venture Partners in 2024; in March 2026, the company reported 259,162 Guardrails AI package downloads in the preceding month.