The llm-typesafe 0.1a0 Plugin Makes TypeSafe AI's Jev Compatible with LLMs

A new plugin, llm-typesafe 0.1a0, brings support for the Jev model from TypeSafe AI to the LLM client. It offers structured responses in yes/no format, predefined choices, or on a scoring scale, complete with ready-to-use examples.

Three modes of structured response, from binary to score

The llm-typesafe 0.1a0 plugin allows the use of the Jev model from TypeSafe AI to obtain structured responses in LLM. It supports "noul" type questions for yes/no answers, with an example output displaying a JSON object containing "type: noul" and a "noul" score of 0.99. The plugin also handles choice questions via the option "-o answer_type choice," with the ability to set guidelines such as prioritizing billing if there are billing and technical issues. The proposed categories include "billing" (charges, invoices, payments, or refunds), "technical" (installation or usage issues), and "other" (no matching category). Finally, it is possible to ask evaluation questions with "-o answer_type score," on a scale ranging from no reproduction instructions to complete steps with expected and actual results.

Usage in LLM: "jev" model, file pipes, and README

The Jev model is invoked in LLM with the option "-m jev." Examples show that the plugin accepts content piped from files, using commands like "cat message.txt" or "cat report.txt" as input. A README is available for additional information on using the plugin.

Installation and TypeSafe API access

The llm-typesafe plugin, version 0.1a0, adds support for the Jev model from TypeSafe AI in LLM. It is installed with the command "llm install llm-typesafe." To use it, one must configure a TypeSafe API key via "llm keys set typesafe." This key is obtained after signing up on a waiting list, which the plugin author indicates is progressing quite quickly.