Datasette Agent
Open-source plugin for Datasette that queries SQLite databases through chat and writes SQL, asking for approval before any change.
What it does
Datasette Agent adds an AI assistant for SQLite databases to the open-source tool Datasette. In the chat you ask about tables and contents in plain language; the agent reads the schemas, writes SQL and runs read-only queries. Typing # attaches specific tables or stored queries. Write statements and saving queries only run after you approve them. Background agents work towards a goal on their own, and there is also a command-line chat. Other plugins can add their own tools, for example for charts.
Who it suits
Teams that already run Datasette and want to analyse data without writing every query themselves. Setting it up requires experience with Python and running servers.
Data protection for businesses
You run the plugin on your own server, so you decide where it is hosted. Schemas and query results are sent to the model provider you configure. Whether inputs are used for training there and which data processing agreement applies is something you settle with that provider. Plugins exist for local models – LM Studio, for example, is documented – but the model must support tool calling. The project offers no business plans and no data processing agreement of its own.
Limits
The plugin is in alpha and requires a pre-release of Datasette 1.0. It only works with SQLite. The model cannot be changed within a conversation. If a write statement fails, earlier statements are not rolled back.