llm-typesafe 0.1a0
Simon Willison’s Weblog Simon Willison ● Covered by 16 sources
Simon Willison released llm-typesafe 0.1a0 for LLM. It adds support for TypeSafe AI’s Jev model and turns vague prompts into structured answers.
Based on reporting by Simon Willison’s Weblog, Simon Willison — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Simon Willison has released llm-typesafe 0.1a0, a new plugin for LLM that adds support for TypeSafe AI’s Jev model. The pitch is simple: instead of hoping an LLM gives you something clean enough to parse, you can ask it to return a specific shape of answer.
Installation is straightforward. First install the plugin with llm install llm-typesafe, then set an API key with llm keys set typesafe. Willison says the key comes from a waitlist, and that the waitlist seems to move pretty fast.
The examples show where Jev fits. A yes-or-no prompt about whether a message explicitly requests a refund returns JSON with a noul field set to 0.99. That is a very LLM-ish way to say “I’m pretty sure,” but in a format a program can use without a lot of pleading.
It also supports choice questions. In one example, the model is told to route a message to billing, technical, or other, with a rule that billing wins if both billing and technical issues appear. Another example asks for a score on how reproducible a bug report is, using a list of criteria that ranges from no reproduction instructions to complete steps with expected and actual results.
Willison points readers to the README for more detail. The release is small, but the direction is clear: less chatty wandering, more structured outputs that can drop straight into a workflow.
My take — AI-written commentary, not fact-checked reporting
This is the sensible part of the AI boom: fewer magic tricks, more answers you can actually route, sort, and trust. The industry keeps selling “agents,” but the real win is getting a model to admit what bucket something belongs in without turning your code into a wrestling match. Structure beats sparkle, every time.
Read more about this at: Simon Willison’s Weblog
