Intel / article

llm-typesafe 0.1a0

Source: simonwillison.net
Date: Published Sep 22, 2026

Context

Simon Willison's -typesafe plugin brings TypeSafe AI's Jev into the LLM command-line tool. Its release examples show text being evaluated against an explicit question: whether a message requests a refund, which team should receive it, or how complete a bug report's reproduction instructions are.

The user supplies the question and any categories or scoring criteria. This makes Jev's structured-decision approach available in command-line workflows: a script can request a probability, category or score instead of interpreting an open-ended written answer.

Key takeaways · AI-distilled

The llm-typesafe plugin exposes answer types for TypeSafe's Jev model beyond yes/no: a 'choice' mode that classifies text against named categories, and a 'score' mode that rates it against an ordered rubric, both set via the -o flag.
A yes/no ('noul') query returns a confidence score rather than a bare boolean, e.g. {"type": "noul", "noul": 0.99} for a refund-request check, per Simon Willison's writeup.
Access to TypeSafe's Jev model requires a waitlist-gated API key that Willison says 'seems to move pretty fast.'

Terms in this piece · Glossary

LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.

Read the source simonwillison.net