SHELFCRITTER AI Digest

TOOL

Datasette Agent

tooltopic-noteagentic-codingstructured-data

Overview

Datasette Agent is an extensible AI assistant for Simon Willison‘s Datasette built on his llm library — conversational SQLite querying with a plugin architecture, released as a first build on 2026-05-21. The live demo runs on Gemini 3.1 Flash-Lite, with a CLI path for local Gemma 4-26B users. The design choice to expose a plugin layer rather than a fixed tool set is a small but pointed bet that the right abstraction for structured-data agents is the data-platform’s own extension API, not a generic tool-calling shell.

Timeline

  • 2026-05-22-AI-Digest — Inaugural release covered in the digest. Useful as a concrete practitioner reference for tool-use agents over structured data; reads alongside the ACC paper (arXiv:2605.21850) the same day — agent trajectories over structured data are now both a training signal (ACC) and a shipping product surface (Datasette Agent) in the same week.

Key Developments

  1. Plugin-First Tool Architecture: Rather than ship a fixed set of SQL/inspection tools, Datasette Agent exposes Datasette’s plugin layer to the model — every Datasette extension becomes a potential tool. The bet is that the data-platform’s existing extension surface is closer to the right abstraction for structured-data agents than a generic tool-calling shell.

  2. Live Demo on Gemini 3.1 Flash-Lite, CLI Path for Local Gemma 4-26B: Shipping the public demo on a hosted Flash-tier model while documenting the local-inference path on a 26B open-weights model is a deliberate two-track posture — hosted for the demo’s accessibility, local for practitioners who want the agent inside their existing Datasette deployment.