Plurai raised $13 million in seed funding and opened its AI-agent simulation platform to all developers, betting that testing agents in realistic virtual environments can prevent failures before customers encounter them.
Team8 led the round, with U&I, Velocity and Mercer participating alongside more than 20 angel investors. Co-founders Ilan Kadar and Elad Levi previously worked on AI systems at Nexar, where simulation was central to autonomous-driving development.
Plurai builds a digital twin of an agent’s environment using a company’s knowledge base, software tools and policies. It then runs thousands of simulated users through normal and adversarial scenarios, searching for hallucinations, policy breaches, broken tool calls and sensitive-data leaks.
When a failure appears, the platform generates fixes and uses the resulting data to train smaller language models for specific tasks. Those models can power agents or continuously evaluate their actions in production, with new failures feeding back into simulation.
Plurai says its design partners reduced failures reaching real users by 82%, cut the cost of operating reliable agents by 97% and shortened some development cycles from months to hours. These are company-reported outcomes, not independent industry benchmarks.
A separate enterprise case study with NVIDIA and Nebius illustrates the economics. A post-trained NVIDIA Nemotron 3.5 Lightning model scored 88.8 out of 100 on production chats, versus 88.6 for GPT-6 Luna and 78.4 for Claude Haiku-5.5. Plurai reports the tailored model cost around 40 times less than GPT-6 Luna and 44 times less than Claude Haiku-5.5 for that specific workload.
The platform is now generally available, including a free tier and support for tools such as Claude Code, Codex and GitHub Copilot. The broader test is whether simulation can make agent reliability measurable and affordable beyond early customers.
Get stories like this in your inbox every week. Subscribe to The Week in Perspective.

