Simon Willison just released Datasette Agent, and if you’ve been following his work on either Datasette or his LLM library, this is the moment those two projects finally collide.

Datasette Agent is exactly what it sounds like: an AI assistant that sits on top of Datasette and lets you ask questions about your data in plain English. It’s conversational, it’s extensible, and it already has plugins that let it generate charts and run sandboxed code.

Willison announced it yesterday after working on his LLM Python library for three years. The timing matters because LLM has become a genuinely useful abstraction for working with language models in Python, and Datasette is one of the better tools for exploring SQLite databases. Combining them was obvious in hindsight, but it took three years to get here.

What it does

The core interaction is simple: you ask a question, the agent queries your Datasette instance, and you get an answer. Add the datasette-agent-charts plugin and it’ll generate visualizations. The 0.1a3 release added “View SQL query” buttons so you can see what queries the agent is actually running, which is the kind of detail that matters when you’re trying to understand if the answers are trustworthy.

There’s also a datasette-agent-sprites plugin that runs commands in a Fly Sprites sandbox. This is interesting because it means the agent can execute code without destroying your system, which opens up more complex data manipulation tasks.

The project is clearly early (everything is alpha), but the foundation is solid. Willison has been building tools for working with data for years, and he understands the difference between a demo and something you’d actually use.

Who should care

If you’re already using Datasette, this is worth trying immediately. It’s a natural extension of what Datasette does well, which is making SQLite databases easy to explore and share.

If you’re building data tools and wondering how to add AI without making it worse, this is a good example. The “View SQL query” buttons are doing real work here. They let you verify what the agent is doing instead of just trusting it, which is the right balance for a tool that’s querying your actual data.

If you’re working with LLM in Python, Datasette Agent is also a useful reference implementation. It shows how to build something extensible on top of the library, with a plugin system that already has working examples.

What’s actually here

The GitHub repo is at github.com/simonw/datasette-agent, and there are already three plugins:

• datasette-agent-charts (0.1a2) for generating visualizations
• datasette-agent-sprites (0.1a0) for sandboxed code execution via Fly
• The core agent itself (0.1a3)

Everything is tagged properly on PyPI, so you can install it and start using it today if you want. Just know that the version numbers mean what they say: this is alpha software.

The documentation shows how to extend it with your own plugins, which is the point. Willison has been building extensible tools for long enough that he knows how to design a plugin API that actually works.

Why this matters

Datasette Agent is interesting because it’s a real tool from someone who builds real tools, not a proof of concept from a research lab. Willison has been publishing databases with Datasette for years. He knows what actually matters when you’re trying to answer questions about data.

The plugin architecture also matters because it means this can grow in useful directions without the core project trying to do everything. Charts are a plugin. Sandboxed execution is a plugin. Whatever comes next will probably also be a plugin.

If you’re looking for examples of how to integrate AI into existing tools without making them worse, this is one to watch. It’s not trying to replace SQL or hide what it’s doing. It’s just making it easier to ask questions and get answers, which is exactly what a conversational interface should do.