Simon Willison released datasette-llm 0.1a11 on 14 November, adding support for arbitrary OpenAI-compatible endpoints to the Datasette data exploration platform.
datasette-llm is the intermediate plugin for Datasette that supports configuring large language models for use with other Datasette plugins Simon Willison1. This release allows users to configure Datasette to use LLMs hosted anywhere that talk the "OpenAI pseudo-standard web protocol" Simon Willison1. The change enables integration with local inference engines such as LM Studio, Ollama, and llama.cpp without requiring specific model plugins.
The update introduces an api_base configuration field within model references, allowing direct connection to any server exposing Chat Completions or Responses APIs compatible with the OpenAI standard. Users can specify whether an endpoint uses Chat Completions or Responses by setting responses: true. The PyPI documentation confirms that the api_base parameter accepts HTTP or HTTPS URLs, including path prefixes such as /v1 PyPI: datasette-llm 0.1a112.
Security considerations remain central to the design. Credentials should be stored in key fields, which resolve through datasette-secrets or LLM's stored keys. The documentation explicitly warns against placing credentials in the headers dictionary, noting that such values could be visible to anonymous users at the /-/config.json endpoint PyPI: datasette-llm 0.1a112. URLs with embedded credentials, query strings or fragments are rejected. An endpoint without a key is treated as unauthenticated and never receives the default OpenAI key.
This release is a pre-release and may not be stable for production use, according to the PyPI page PyPI: datasette-llm 0.1a112. Independent testing of the OpenAI-compatible endpoint implementation has not been established by external security researchers or benchmark labs. Installation requires installing the plugin in the same environment as Datasette; while general installation requires at least one LLM model plugin, the new api_base feature requires no additional LLM plugin or changes to LLM's model configuration file PyPI: datasette-llm 0.1a112. The 0.1a10 release notes were unavailable, so changes relative to earlier versions could not be confirmed.