AI-rewritten: This is a summary of an article from Tom’s Hardware, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article – read it for the full story.

---

TypeSafe recently launched Jev, an API-driven system that uses decision models to answer specific questions with probabilities rather than generating text. This approach avoids the usual issues of large language models, such as conversational hallucinations or the need for extensive context history. Instead of writing out answers word-by-word, the model provides direct data in the form of percentages that can be used in code "if" statements. The technology is designed to be fast and inexpensive compared to standard AI chatbots.

Nandakishor Mukkunnoth, founder of ConvAI Innovations, stated he developed the base architecture for Jev over a year prior to TypeSafe’s release. He published his original work on March 30, 2025, in an arXiv paper titled SalesRLAgent, which included model weights and datasets. Mukkunnoth noted that his initial design was intended for a specific sales use case, whereas TypeSafe expanded it into a general-purpose engine. Despite feeling overlooked initially, he chose not to remain bitter and released his own open-source version called Laya on the same day as Jev.

The rapid release of Jev sparked significant interest, leading many companies to create similar services within weeks. Notable examples include Microsoft Decision, OpenAI Decisions API, and Fastino Glide. Several other projects like Amazon’s Strands Decider 2B and Cloudflare Clef also emerged with open weights for local use. Most of these implementations rely on lightweight variants of Alibaba’s Qwen model, though Mukkunnoth’s Laya uses encoder-only models from the Bert family. The term "System One" used by both TypeSafe and Mukkunnoth references Daniel Kahneman’s book Thinking, Fast and Slow to describe a faster mode of thought.