Datasette-llm 0.1a11 adds support for OpenAI-compatible API for local models
The datasette-llm plugin for the Datasette tool has been released in version 0.1a11 (alpha) with support for OpenAI-compatible API endpoints, making it possible to connect local models via LM Studio, Ollama, or llama.cpp.
Datasette-llm is an intermediary plugin for the Datasette tool, used to configure LLM models for other Datasette plugins. The new version 0.1a11 adds support for OpenAI-compatible API endpoints — that is, the web protocol used by OpenAI, which is also mimicked by a number of other model-hosting tools.
Thanks to this, Datasette can be configured to use a model running anywhere that, according to the source, speaks this pseudo-standard protocol, including locally running models via LM Studio, Ollama, or llama.cpp (which, according to the source, has just reached version 1.0.0). The source gives a specific example of running it via uvx without installation, where you set the model, its name, and the address of the local API (e.g. http://localhost:1234/v1), resulting in a locally running instance of Datasette Agent powered by your own model.
This is an alpha version (0.1a11), published by author Simon Willison.
Why it matters
It allows developers and analysts who use Datasette for working with data to replace a cloud API with a locally running model — without having to pay for tokens and without sending data outside their own machine or network. This is relevant especially for those who already work with Datasette or are considering deploying Datasette Agent.
Two audiences, two different impacts
What this means
For individuals
Developers and analysts using Datasette can now easily connect a locally running model (LM Studio, Ollama, llama.cpp) instead of a paid cloud API.
For a business
Companies using Datasette for data analysis can deploy Datasette Agent with a local model, thereby reducing cloud API costs and preventing sensitive data from being sent outside their own infrastructure.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.