Simon Willison released version 0.1a0 of llm-typesafe, a plugin for the LLM command-line tool that adds support for TypeSafe AI’s new Jev model. Users install it via a single command and set an API key to enable specific question types. The tool accepts yes/no queries, such as determining if a message requests a refund, and returns a numerical score like 0.99. It also handles choice questions where the model selects an option from defined categories, such as routing a ticket to billing or technical support. Scoring questions allow users to input text and receive a structured rating based on provided criteria, for instance assessing the reproducibility of a bug report.
This update addresses a specific gap in how developers interact with large language models by enforcing strict output formats. Instead of relying on the model to guess a format, the plugin forces a specific answer type like choice or score. This reduces the need for complex prompt engineering and makes the results easier to parse automatically within scripts. The feature set expands the practical utility of the command-line interface for tasks requiring deterministic responses.
* Supports yes/no, multiple choice, and scoring question types
* Returns JSON output for easy programmatic parsing
* Requires an API key from TypeSafe AI
