TypeSafe introduced the Jev decision-making API

Company TypeSafe released an API for Jev — a tool that instead of generating text selects the optimal option from a predefined solution space. The system is focused on tasks that require high speed and accuracy of execution, excluding hallucinations of standard language models.

The Jev architecture is based on Noul, Choice, and Score contracts, which define the logic of choosing a response. Unlike classical LLMs, the model does not engage in predicting the next token, but performs classification or action selection based on given parameters. This allows for achieving extremely low latency: for example, a browser-based agent based on Jev can find airline tickets in 7 seconds.

Developers are already using the API to create trading bots, content filters, and management systems, including prototypes for Tesla FSD. The tool allows developers to strictly limit the scope of acceptable responses, making the system's behavior predictable in critical scenarios. The documentation describes the integration process through API calls, where cost and performance are optimized for specific applied tasks.

The transition from generative models to specialized solvers allows reducing computational costs and increasing the reliability of automated systems. Jev fills a niche where a creative approach is not required, but rather strict adherence to predefined decision-making algorithms in real-time.