Guide demonstrates typed judgments with TypeSafe AI’s Jev model
A new coding tutorial presents Jev, TypeSafe AI’s first System One model, as a tool for structured decisions rather than text generation. Applications provide program state together with typed questions, and receive outputs that can be used directly in code branches. The guide shows installation of the official Python SDK and a request combining Choice, Score and Noul primitives.
The tutorial examines how different state formats affect what Jev can determine, including strings, arrays and structured objects. It also explains how to recompute the published confidence statistic from returned probabilities and compares sending ten questions in one batch with making ten separate requests.
Further examples cover confidence-gated routing, composite scores whose weights remain in application code, typed function calling and counting approaches suited to the model. The production-oriented section uses Pydantic response models, an asynchronous client with asyncio fan-out, retry policies and typed errors. A running ledger records calls, token usage and the cost of the notebook, with Jev’s listed input pricing applied.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.