1min.

1MIN · PYTHON

Datasette LLM adds direct connections to compatible model servers

Datasette LLM 0.1a11 lets other Datasette plugins use a model server through an OpenAI-compatible endpoint without installing an additional model plugin. Configuration separates a local model alias from the remote model name and API address. This extends model selection to local servers as well as hosted providers, while applying model allowlists and credential checks more consistently.

Datasette · Nov 14, 2026

Why it matters

Plugins can change inference providers through configuration while retaining explicit control over available models and credentials.

More context

Requires LLM 0.37.1 or newer. Configure an endpoint or default-model dictionary with model, model_name and api_base; test a local LM Studio, Ollama or llama.cpp server. Put credentials in the key setting, not extra headers, which are visible in Datasette configuration output.

Keep in mind

Alpha release with compatibility changes: excluded models and missing named credentials now fail earlier. Server protocol and tool support still need testing. Local inference needs suitable hardware; hosted endpoints retain their own pricing and data policies.

Original read

More engineering ideas, one minute at a time.

Explore more stories