- Published on
Datasette-llm 0.1a9 adopts LLM default model for Datasette Agent
- Authors
- Name
- DeepHorus
Datasette Agent users can now begin work without first naming a model, after the datasette-llm plugin was updated to version 0.1a9 on 31 October 2026. The release makes Datasette Agent inherit the default model already used by LLM. The change targets a small but persistent setup friction: an extra configuration step that previously stood between a user and a first run.
Datasette Agent runs on datasette-llm, the plugin that connects Datasette's data exploration environment to language models. Until now, users had to explicitly set a default model for the agent. That requirement meant even a quick trial depended on knowing and supplying a model identifier.
With 0.1a9, a basic launch can be performed with uvx --with datasette-agent datasette content.db --internal internal.db. Previously, the same command also needed -s plugins.datasette-agent.default_model gpt-6-astra. The new behaviour removes that explicit flag by falling back to LLM's configured default model.
The adjustment lowers the barrier to entry for new users and makes demos, documentation examples, and ad hoc experiments shorter. It also reduces the chance of configuration errors before an agent session starts. Existing workflows that specify a model explicitly can continue to do so.
The update is described as a tiny user experience improvement rather than a feature overhaul. Its significance lies in defaults: tools become easier to adopt when they work with sensible existing settings. For Datasette Agent, that means less setup prose and more immediate interaction with data.
More substantial changes are reportedly in the pipeline for Datasette Agent. The 0.1a9 release therefore serves as a stepping stone while broader upgrades are prepared. For now, users can launch the agent with fewer flags and rely on LLM's default model selection.
