the Gemini 2.5 Pro’s enhancement is obvious. I wish to use it early in DT4.The problem maybe easy for you exciting team.
And the Deepseek API method is also useful.could DT4 add this?
Thank You!
This is indeed planned but currently Gemini 2.5 Pro is still experimental.
Is it the policy of DEVONthink not to give access to experimental models? Why, if I may ask?
Gemini 2.5 Pro is really good for tasks that require long context… such as working with multiple documents in a DEVONthink database.
Note DEVONthink 4 is not “an AI application”. External AI is not the core of the application. It is a feature complementary to DEVONthink’s AI, not a replacement. And it is not the focus of development to try and compete with bespoke AI applications.
1 Like
Interesting response…
Gemini 2.5 Pro (while indeed still experimental) is available via the Gemini API, which is available in DT4. So it seems fair to ask whether we can hope for this model to become accessible via DEVONthink.
Experimental models are usually, well, unreliable and therefore could cause additional support and/or a poor user experience.
I see that reasoning. However, experimental models also tend to be cutting-edge. So I guess it’s a tradeoff between giving power users what they want and keeping support requests in check 
Sooner or later we’ll definitely add it.
Is it planned or conceivable that DT4 will receive a connection to locally operated LLM (e.g., Ollama for Docker)?
You should read the Getting Started > AI Explained section of the built-in Help and manual.
I am very happy with the AI features added to DEVONthink 4, they have helped me very well. Thanks for your hard work!
I’m wondering if the team will update the AI options frequently now that they are faced with such rapid development of AI model updates? Or maybe it might be more convenient to provide a customized option? The recently updated Gemini 2.5 Pro and Grok 3 are both models I’ve been really enjoying lately and would love to be able to use them soon.
1 Like
We prefer to maintain a carefully selected list of models but Gemini 2.5 Pro will be supported by the next beta.
3 Likes
Yes. Grok 3 is wildly better for the tasks I’ve been giving them. I especially like that it actually gets better with iterative prompting but Gemini and Claude tend to work themselves into rabbit holes or dead ends