I’m trying to rename documents with my local olama running instances. I remember it worked with lower version of 4.x .
When i start applying the rule , no reaction.
I tried also with ChatGPT API key but no response ,too .
As an alternative workaround i have currently a script that works with local olama and does the correct thing. At some time the native one broke for me.
I have DEVONthink 4.4.1 and lates MacOS 26.6.2
Do the PDF documents have a text layer? In that case their kind should be PDF+Text.
In DEVONthink, hold the Option key and select Help > Report Bug and we can look at the information and logs.
Yes they have, and it works with external script. All docs are OCR before they arrive in Inbox:)
I send a bug report. Additional when I use chat assistant the prompt goes through but it doesn’t seem to see the text. Because summarize don’t work.
I can then see the post in ollama log , but i assume also content is not getting there.
It seems to work at least with Chat Assistent when i use ChatGPT
Some more findings:
localhost:11434/v1). The model loads, but same result.plainText() via script returns the full text.Looks like PDF content isn’t passed as text to local providers, maybe sent as a file attachment that only cloud models can handle?
DEVONthink never uploads raw documents to cloud models, that’s part of its privacy strategy. But I wonder whether the document, one of its enclosing groups or tags or the complete database are excluded from MCP & chat?
A more precise prompt might also help in case of small local models. E.g “selected content” is ambiguous, “selected document” is better.
Same default prompt works with chatgpt in chat. I’m just switching the models and it works. It seems only local is not working for some reason. Markdown as described with same chat assistant is working.
Did you try a more precise prompt as suggested?