Datasette Agent: First Release of an Extensible AI Assistant for Datasette
Install Datasette Agent, add the three plugins, set the default model to Gemini 3.1 Flash‑Lite, and test queries against example databases.
Install Datasette Agent, add the three plugins, set the default model to Gemini 3.1 Flash‑Lite, and run a test query against the global‑power‑plants database.
Summary
Datasette Agent has just launched, adding a conversational interface that lets users ask natural‑language questions of SQLite databases stored in Datasette. The release ships with three core plugins: datasette-agent-charts, which renders charts via Observable Plot; datasette-agent-openai-imagegen, which calls ChatGPT Images 2.0 to generate images; and datasette-agent-sprites, which executes code in a Fly Sprites sandbox.
The demo runs on Gemini 3.1 Flash‑Lite, a low‑cost, fast model that can generate SQL queries on the fly. Users can also run the agent against local models such as gemma‑4‑26b‑a4b in LM Studio using a simple uv one‑liner. Future plans include a stable release of the LLM 0.32a0 refactor and integration with Datasette Cloud.
The project is extensible, encouraging developers to write new plugins with Claude or OpenAI Codex, and it already powers a personal assistant called Claw that pulls data from various digital life sources.
Key changes
- Datasette Agent provides a conversational interface for querying SQLite data
- Gemini 3.1 Flash‑Lite is the default model for fast, cheap SQL generation
- datasette-agent-charts adds Observable Plot chart rendering
- datasette-agent-openai-imagegen enables ChatGPT Images 2.0 image generation
- datasette-agent-sprites allows code execution in a Fly Sprites sandbox
- Local models such as gemma‑4‑26b‑a4b can be used via LM Studio with a uv one‑liner
- Future stable release of LLM 0.32a0 refactor and Datasette Cloud integration