LLM plugin adds TypeSafe's Jev model for yes/no, category and score answers

single source· 1 articles · confidence: medium · first seen 2026-09-22 15:54 UTC

What this means for you

If you already drive LLM from scripts, this returns a typed answer — yes/no, a category, a score — instead of prose you have to parse. Access needs a TypeSafe key, waitlisted as of the post. Otherwise nothing to do: the package is at 0.1a0.

Simon Willison released llm-typesafe 0.1a0 on 22 September, a plugin that adds TypeSafe AI's Jev model to LLM, his command-line tool for sending prompts to models. Install it with `llm install llm-typesafe`, then set a TypeSafe API key. Instead of prose, answers come back structured: a yes/no verdict with a confidence score (the plugin calls that answer type "noul"), one of a set of named categories, or a score chosen from a list of criteria. A yes/no call returns, for example, {"type": "noul", "noul": 0.99}. Willison says the waitlist for a TypeSafe key seems to move fairly fast; the post gives no pricing or evaluation figures.

Key facts

  • ·llm-typesafe 0.1a0 was published on 22 September 2026 by Simon Willison. source
  • ·The package is a plugin for LLM, Willison's command-line tool for sending prompts to models, and adds support for TypeSafe AI's Jev model. source
  • ·It installs with `llm install llm-typesafe` and uses a key set via `llm keys set typesafe`. source
  • ·It supports three answer types: noul (yes/no), choice, and score. source
  • ·A yes/no call returns output of the form {"type": "noul", "noul": 0.99}. source
  • ·Willison writes that the TypeSafe API key waitlist seems to move fairly fast; no price is given. source

What the sources say

  • Simon Willison — Release note for a plugin routing LLM prompts to TypeSafe's Jev model, with usage examples.

Sources

The original reporting. Follow these — they did the work.

← the wire