New York-based startup Plurai has raised $13 million to develop a platform for testing AI agents before they are deployed to users. Its engine subjects these agents to thousands of synthetic, multi-turn scenarios to identify failures, while its “vibe-training” process turns rules written in natural language into compact, specialized models. The company claims to reduce evaluation failures by more than 43% compared with GPT-5-mini, operate at an eightfold lower cost and deliver up to 15 times greater coverage of edge cases. Founded by Ilan Kadar and Elad Levi, Plurai has also released IntellAgent, an open-source framework for multi-agent simulation, and is working with NVIDIA on its Nemotron models and NIM software.
Source: Read the original article
