Plurai Raises $13 Million to Simulate Enterprise AI Agent Tasks

Plurai raised $13 million in seed financing to build a system that lets businesses rehearse AI agents inside replicas of their own software, aiming to expose weak spots in workflows that cross company tools.

Team8 led the round, with U&I, Velocity and Mercer participating. Plurai announced the financing Nov. 11.

The platform builds a digital replica of a company’s electronic environment, including its knowledge base and the tools available to its agents. It maps possible task paths, then uses those scenarios to evaluate agent behavior and address edge cases.

Examples include an agent booking travel outside a company’s spending rules, struggling to combine information from multiple datasets or losing track of a long customer service conversation. Plurai’s system is designed to help companies examine those failure patterns through simulation.

The platform works alongside tools such as Claude Code, Codex, Copilot and Cursor. Plurai’s co-founders, Ilan Kadar and Elad Levi, previously worked at Nexar, an autonomous-driving company. Kadar also worked with Waymo on simulated scenarios and evaluations of its driving software.

Plurai’s longer-term plan is to use simulation data to post-train small, open-weight models tailored to individual customers. Tests conducted with Nvidia and Nebius found Plurai’s models performed better than midtier frontier models at lower cost. Kadar said the approach could deliver “an order or two orders of magnitude cost reduction.”

Plurai’s focus on training models for agent tasks follows other work in the field. Oumi recently extended its platform to agent development and model training.