Support for Apple Foundation Models

I was wondering if DT4 will add support for the on-device Foundation Models announced by Apple at WWDC25 that will be included as part of macOS 26.

To date, I haven’t experimented much with AI overall and was not particularly excited by the AI additions to DT4 for myself. I also have not had the time to try the DT4 beta.

I have read the very well-written DT4 documentation about AI, how it is integrated into DT4, as well as a number of forum threads about it.

The AI features in DT4 seem very thoughtfully considered, well done, and flexible in terms of models.

Being able to use the on-device Foundation Models in DT would potentially allow a user to get started using the AI features in DT4 without needing to sign-up for a separate account, pay for a model subscription, send info off to a server or download a local model and keep it up to date.

I am not informed enough about AI model capabilities to know if the Foundation Models would be a good fit for DT4, but if so, using them could make the AI features of DT4 useful “out of the box”.

I know everyone working on DT4 (devs, docs, etc.) has plenty to do to finish DT4 and ship it. Thank you all for your efforts!

Thanks for the kind comments and encouragement. It’s very appreciated!

So far the Foundation Models are much more limited than you may expect, so while it may be possible to add, the actual usefulness within the context of DEVONthink is up in the air. Logically Apple will improve things over time but (1) I wouldn’t expect massive improvements by the time they release Tahoe and (2) the scope of the functions still may not provide any real benefit in our app.

1 Like

We will definitely investigate this. However, a good open-source model (e.g. via Ollama) is most likely still the better option. Of course depending on the hardware.

1 Like