datasette-llm 0.1a11 Release: Support for Any OpenAI-Compatible Endpoint in SQLite DatabasesBrief · digest for November 15
Brief≈ 1 min

datasette-llm 0.1a11 Release: Support for Any OpenAI-Compatible Endpoint in SQLite Databases

Simon Willison just released datasette-llm 0.1a11 for Datasette, the open-source tool for exploring and analyzing SQLite databases right in your browser. The plugin acts as a unified adapter, letting other plugins in the ecosystem talk to language models through a single interface.

The headline addition is support for any OpenAI-compatible endpoint—the shared web protocol spoken by almost every modern model server. What I love most here is the privacy payoff: you can now query sensitive tables without sending a byte to external clouds by hooking up a local model through LM Studio, Ollama, or the new llama.cpp v1.0.0.

In his release note, Willison showed how to spin up the whole stack in one uvx command with zero manual package installs. Simply launch datasette alongside datasette-agent and pass -s plugins.datasette-llm.default_model pointing to your local http://localhost:1234/v1 endpoint and a model like qwen/qwen3.8-27b (lmstudio-qwen). A local server boots straight into an autonomous agent that queries content.db and answers questions about your tables completely offline.

Just remember the plugin is still in alpha: for complex schemas, your local model needs reliable tool calling and clean JSON output, and compact models under 7B parameters can trip over trickier SQL dialects.

Want a deep diveVotes help us choose what to break down in detail. This material has 0 votes.
Sources · 2