@xezpeletaXabi yes, using a local LLM works with datasette-extract already

I'd suggest using Ollama to run a model like Gemma 3 or Mistral Small 3.1 (though others could work well too) - with the llm-ollama plugin installed datasette-extract should work fine

I don't know of any local models that handle PDFs in Ollama yet but plenty of them can handle images