datasette-llm 0.1a9 now inherits the default model configured in the LLM CLI, so Datasette Agent runs straight from one command without a hardcoded model flag. For small teams running database agents, this removes a config step and makes swapping models - cheaper, local, or private - a one-line change.
Simon Willison shipped the release on 31 October. Previously you had to pass -s plugins.datasette-agent.default_model gpt-6-astra every time; now uvx --with datasette-agent datasette content.db --internal internal.db just works, picking up whatever default LLM has already set.
What it means for small business automation
If you want a natural-language layer over a SQLite reporting database - sales, bookings, inventory snapshots - Datasette Agent is one of the simplest routes. The default-model pickup matters because it decouples the agent from any specific vendor: prototype with a frontier model, then switch to a local Llama or Claude Haiku for production without touching the agent code. For a shop with privacy constraints or a tight budget, that is the difference between a toy and a system you can actually deploy on your own data.
Why it matters: Model routing is becoming boring plumbing, and that is good news - your reporting agent should not care which brain is answering, only that the SQL comes back correct.
Source: Simon Willison