HeadsUpAI

Simon Willison Releases Datasette Agent to Query SQLite Databases with Natural Language

Simon Willison released the first alpha of Datasette Agent, a conversational assistant for Datasette that turns SQLite databases into interactive chat interfaces. By integrating his refactored LLM library, Willison demonstrated how models use tool calling to autonomously reason about database schemas and generate SQL.
Default demo model
Gemini 3.1 Flash-Lite
Supported databases
SQLite
Initial plugins
Charts, OpenAI Imagegen, Fly Sprites
Deployment options
Local (uv), Web demo, Datasette Cloud
Model compatibility
OpenAI, Anthropic, Gemini, and others

This release marks a shift from simple text-to-SQL prompts toward autonomous data agents that can iterate on results. While the live demo uses gemini-3.1-flash-lite for its speed and low cost, the architecture is model-agnostic. This allows developers to swap between frontier APIs and local open-weight models based on privacy requirements.

The agent is extensible through plugins like datasette-agent-charts, creating a foundation for personal AI assistants built around private data. The project signals a move toward AI-native data tools where the assistant is the primary operator. You can run the alpha locally via uv or try the interactive web demo.

View the full update on simonwillison.net
Simon Willison
Simon Willison
@simonw
X

I released the first alpha of Datasette Agent - a conversational AI assistant for Datasette that can answer questions about data in SQLite databases, and can be extended with plugins to add extra tools and features Here's a demo https://t.co/2gyduf5Eph

3retweets23likes
View on X

Still wondering? A few quick answers below.

Datasette Agent is an open source conversational AI assistant designed to interact with SQLite databases. It provides a natural language interface where users can ask questions about their data. The agent reasons about the database schema, generates the necessary SQL queries, executes them, and then explains the results to the user.

Yes, Datasette Agent is an open source project. You can run it locally using the uv tool with a single command that specifies your database and preferred AI model. It is also available as a live interactive demo at agent.datasette.io, though this version requires a GitHub sign-in to prevent service abuse.

The agent is model-agnostic because it is built on the LLM Python library. It supports hundreds of different models, including frontier APIs like GPT-5.5, Anthropic Claude, and Google Gemini. It also works with local open-weight models such as Qwen 3.5 or Gemma 4 when served through tools like LM Studio.

Datasette Agent uses a plugin architecture that allows developers to add custom tools and capabilities. At launch, three plugins are available for generating data visualizations using Observable Plot, creating images via ChatGPT Images 2.0, and executing code within a persistent Fly Sprites sandbox. Developers can build additional plugins to further customize the agent.

To function effectively, the chosen AI model must support reliable tool calling, which is the ability for a model to invoke external functions. Specifically, the model needs to be proficient at generating valid SQLite queries that can run against the connected database. Many recent open-weight and frontier models are now capable of handling these tasks.

Share this update