Simon Willison Releases `llm-typesafe` Plugin for TypeSafe AI's Jev Model
Simon Willison announced `llm-typesafe` 0.1a0, an alpha plugin for his `llm` CLI tool that adds integration with TypeSafe AI's new Jev model. The plugin allows command-line users to query Jev directly for structured, typed decisions such as yes/no probabilities, category selections, and numerical ratings. This plugin expands the utility of the popular `llm` CLI tool to non-generative decision models, enabling developers to run lightweight structured evaluations directly within terminal scripts. It highlights an emerging category of specialized AI models optimized for fast, calibrated decision-making rather than traditional free-form text generation. Installed via `llm install llm-typesafe`, the plugin requires a TypeSafe AI API key to function. It supports specific Jev response modes including `noul` (yes/no probability questions), `choice` (categorizing inputs against criteria), and `score` (evaluating inputs against structured rubrics).
## BACKGROUND
TypeSafe AI's Jev is a specialized AI model designed for high-speed, typed decision-making rather than open-ended natural language generation. Instead of outputting standard text tokens, Jev evaluates prompts against defined criteria and returns structured probabilities and categorical responses. Simon Willison's `llm` is an open-source command-line application and Python library designed to interact with diverse language models through extensible plugins.