datasette-llm 0.1a7

The meaningful detail here isn't temperature defaults specifically, it's that datasette-llm is now absorbing configuration concerns that previously lived at the query or script level, which is the kind of API surface decision that either attracts serious integrators or quietly limits the plugin to power-user hobbyist use.

Willison's iNaturalist project from early May (covered here) showed him using LLM-assisted tooling to build real data workflows under real constraints, and datasette-llm is the more formal, shareable expression of that same impulse. The broader pattern across recent coverage is a push toward structured, inspectable AI workflows rather than single-shot inference, visible in the chart generation validation pipeline piece and in RunAgent's constraint-guided execution work. Datasette-llm's model-level defaults are a small but concrete step in that same direction: reducing the surface area where ad-hoc configuration can introduce inconsistency across a pipeline.

Watch whether datasette-enrichments-llm adopts these configuration defaults as a first-class dependency in the next two or three releases. If it does, that confirms the plugin pair is converging into a coherent platform layer rather than staying loosely coupled experiments.