Version 1

Simon Willison launched Datasette Agent on May 21, 2026, a conversational natural-language SQL interface built on top of his Datasette data-publishing tool [1]. He describes the project as the convergence of his two major open-source efforts—the LLM library and Datasette—after three years of parallel development [1]. The live demo runs on Gemini 3.1 Flash-Lite, and three plugins already ship with the system: charts, image generation, and a Fly Sprites persistent sandbox executor [1][2]. Willison has been iterating rapidly, releasing multiple alpha versions of the core package and chart plugin in the 48 hours surrounding the announcement [4][3].

Datasette Agent demonstrates a practical, plugin-based pattern for bolting LLM-powered natural language interfaces onto existing data infrastructure—without replacing the underlying tool. Willison's observation that open-weight models released in the past six months now reliably generate SQLite queries and tool calls [1] suggests the launch timing is tied to a genuine capability inflection, not just a feature milestone. His claim that Claude Code and Codex can autonomously write new plugins from a reference repo [1] points toward a self-extending ecosystem that could grow faster than a single maintainer could sustain alone.

On May 21, 2026, Simon Willison published the central announcement for Datasette Agent, framing it as the moment his two flagship open-source projects—the LLM library and the SQLite-based data publishing tool Datasette—finally converged after three years of parallel development [1]. The result is a chat-style interface that lets users query any Datasette database in plain English, with the agent translating natural language into SQL behind the scenes and returning structured results.

The architecture is deliberately extensible. Three plugins shipped in rapid succession around the announcement: a charting plugin (datasette-agent-charts), an image generation plugin, and datasette-agent-sprites, which allows the agent to execute arbitrary commands inside a Fly Sprites persistent sandbox [1][2]. Willison notes that Claude Code and OpenAI Codex can both write new plugins effectively when given the reference repository as context, lowering the contribution barrier considerably [1]. The live public demo uses Gemini 3.1 Flash-Lite, chosen for its speed, low cost, and reliability on SQLite query generation [1].

Willison also made a broader observation about the model landscape: open-weight models released in the past six months are increasingly capable of reliable tool calls and SQL generation against SQLite [1]. This positions Datasette Agent's launch as deliberately timed to a capability inflection rather than as a proof-of-concept ahead of its time.

The release cadence in the days surrounding the announcement has been fast. The datasette-agent-charts plugin (version 0.1a1, May 20) added magnitude-based sequential color shading for bar and waffle charts, interactive tooltips, and—notably—a permission check that now verifies execute-sql access before running the column-name discovery query [3]. The core datasette-agent package reached 0.1a3 on May 21, adding 'View SQL query' buttons for both visible tables and collapsed tool-call results, suppressing empty reasoning chunks from the UI, and improving graceful handling of truncated SQL responses so result tables still render even when the agent's output was cut short [4].