Ollama local model not receiving document content in Chat

I have DEVONthink 4 with Ollama (qwen3:14b) configured at http://localhost:11434/api/chat/. The connection works fine — the model responds to direct questions and the Context Window shows token usage correctly.

The problem is that when I open a document and start a Chat, the document content is never passed to the model. The model responds asking me to “provide the text you want to analyze.”

I have tried:

  • Documents set to Auto and Text — same result

  • Manually pasting the document content into the chat works perfectly

This confirms Ollama itself is working, but DEVONthink is not injecting the document content into the conversation automatically.

Is this a known limitation with local Ollama models, or is there a setting I’m missing?

How did your prompt look like and what kind of document did you use? Especially in case of smaller and dumber local models the prompt has to be more precise frequently.

That worked! Using “list the key points of the text” did nothing, but switching to “Please read the currently selected document and list its key points” worked perfectly. Didn’t expect the wording to make such a difference!

Are there other prompt patterns I should keep in mind when working with local models in DEVONthink? Any tips would be appreciated. Thanks!

Explicitly referencing the document(s) or the selection is basically the only one, especially in case of local models. Apart from that the usual prompting recommendations matter but this isn’t specific for DEVONthink.