Simon Willison released datasette-llm 0.1a9, an LLM integration plugin that other Datasette plugins depend on. According to his weblog post of 31 October 2026, it now uses the default model configured in his LLM tool, so Datasette Agent can start without a separately configured default model.
What does the 0.1a9 release change?
The release makes one small change: datasette-llm reads the default model from LLM. Simon Willison's Weblog calls it a tiny user experience improvement to the underlying plugin, not a new capability.
Previously, according to the post, users also had to pass a model to Datasette Agent with an option such as -s plugins.datasette-agent.default_model gpt-6-astra. That extra step is no longer needed if LLM already has a default model.
The author says bigger upgrades to Datasette Agent are in the pipeline, but the post does not describe them. It is a short note, not a full changelog.
How can you try it and check the change?
The post gives no install steps beyond the example above, so the practical first step is to look at the datasette-llm repository and compare its release notes and commits with your setup.
- Confirm LLM already has a default model set.
- Run Datasette Agent without the plugin-specific default model option.
- Keep your explicit setting until you have confirmed the default is picked up.
Because datasette-llm is meant as a shared layer, other plugins that depend on it may also benefit. The source does not say which ones.
Who it is for and where it falls short
This release suits Datasette users who already use LLM and want fewer setup steps. The version string 0.1a9 marks an early alpha, so behaviour and interfaces may still change.
The source does not state which LLM versions are supported, which model types can serve as the default, or what happens when LLM has no default model. The behaviour is described by the author and has not been independently verified.
According to GitHub as of 1 November 2026, the repository has 20 stars and 1 fork, is licensed Apache-2.0, is written in Python, and had its last commit on 31 October 2026. That points to a small project with recent activity, but says nothing about production use.
Bottom line: If you already run LLM with Datasette Agent, this release removes one configuration step and is worth trying in a non-production setup first. If you need a stable, widely used plugin, wait for a later release.
Source and verification: The information comes from Simon Willison's Weblog, in a post dated 31 October 2026 and listed as published on 1 November 2026 (source post). Repository figures come from GitHub as of 1 November 2026. This article has not run the release.
Related: DoorDash says only several hundred orders came from AI agents · Zest's AI Restaurant Concierge Now Books Tables Over Text
Comments (0)