What decision models are
Jev and its clones are APIs used directly inside your code. You feed them input data such as a customer's transactions along with questions: "is the user looking for a return?" for a yes/no answer, or "which item are they referring to?" for a multiple-choice answer. The model returns a probability for each question, which you then use in your own if-statements.
This sidesteps the usual problems of coaxing an LLM into a direct answer. There is no token-by-token generation, no conversation history, no context stuffing and no conversational hallucination. It is also fast, answering questions in parallel, and far cheaper than standard LLMs.
Who copied whom
TypeSafe released Jev alongside a brief description of its System One architecture. Days later, ConvAI Innovations founder Nandakishor Mukkunnoth wrote that he had designed the Jev architecture a year earlier. He pointed to his arXiv paper on SalesRLAgent, published March 30, 2025, along with his model weights and dataset, and to a September paper that more closely matches how Jev operates.
Mukkunnoth said he was overlooked because his idea was built for a vertical use case. TypeSafe shipped Jev as a complete hosted product with servers, an API, documentation, developer kits and pricing. Mukkunnoth released the open-source Laya model, which runs on your own hardware with multi-language support, around the same day.
The market fills up fast
Within weeks of Jev's release came Microsoft Decision, OpenAI Decisions API and Fastino Glide. Cloudflare Clef and Perplexity Decider arrived with open weights, meaning you can run them on your own gear. On the open-source side there is Amazon's Strands Decider 2B plus Kev, Von, Jeff, jevlike, SemIf, OpenJev, NanoJev, Ollaya and many others.
Most of these projects share a common root: taking an existing lightweight model, usually a variant of Alibaba's open-source Qwen, and adding layers that shape input, output and decision-making. Mukkunnoth's Laya instead uses encoder-only models from the Bert family.
Why it matters
Decision models pull AI out of the chat interface and embed it inside software. That cuts both cost and latency, while raising the question of who came up with the idea first.
