TypeSafe’s Jev sparks copycat decision models after researcher claims prior work

Nandakishor Mukkunnoth, founder of ConvAI Innovations, accused TypeSafe of copying his earlier research. Mukkunnoth published an arXiv paper on SalesRLAgent in March 2025, detailing a probability-estimation model and open-source weights. He later refined the idea in a September paper, aligning closer to Jev’s architecture. Frustrated by what he called “vertical use-case neglect,” he released Laya, an open-source Jev-like model with multi-language support, on the same day TypeSafe launched Jev. Both models reference System One from Kahneman’s Thinking, Fast and Slow, though their implementations differ—TypeSafe’s is hosted, while Laya runs on user hardware using Bert-family encoder-only models.

Within weeks of Jev’s debut, major players rushed alternatives: Microsoft’s Decision, OpenAI’s Decisions API, and Fastino’s Glide. Open-source projects like Strands Decider 2B, Kev, Von, Jeff, jevlike, SemIf, OpenJev, NanoJev, and Ollaya emerged, often repurposing lightweight models like Alibaba’s Qwen. Cloudflare’s Clef and Perplexity’s Decider also joined, offering open weights for self-hosting.