A.I. integration?

[Jon]

Now that macOS 15.2 is available, you will find that Apple Intelligence writing tools work in Bookends now for the PDF reader and Bookends Browser (to use it, you must have a compatible Mac and be in a country/region where Apple Intelligence has been enabled). More will be more coming in future updates.

Jon
Sonny Software

Addendum. Edit -> Writing Tools is also available when the a field in the edit pane has the focus.

[vinschger]

Dear Jon,

As far as I am aware, Apple's AI is currently not available outside of the United States. Do you plan to integrate the OpenAI or Claude API in the future, allowing users to provide the license keys for these services? Or do you not see significant benefits in supporting other AI providers in terms of features and potential?

This is merely a question and not a feature request on my part, as I am not in a position to assess whether such an integration would truly benefit users.

What do other users, particularly those with experience using different AI providers or services, think? Does the choice of provider make a meaningful difference?

[Jon]

Apple Intelligence will of course be coming worldwide (the goal is April, I believe). And in fact you can use it anywhere now if you set your system language and Siri to English (probably US) and perhaps another tweak or two. Google for details and discussions.

Jon
Sonny Software

[iandol]

What do other users, particularly those with experience using different AI providers or services, think? Does the choice of provider make a meaningful difference?

Not a meaningful difference to be honest. I've used Claude and the recent latest Gemini and also regularly run local models (Llama 3.2 as a fast model and IBM granite 3.1 as a bigger model). Local models should in theory be much less powerful, but honestly for things like summarising and text revision, local models are close to the "big" foundation models practically. I use LM Studio to run these on a macbook air and integrate them into the OS with BetterTouchTool and an Alfred workflow called NexusGPT, so I can do most of what Apple intelligence does locally while we wait for it (and I will probably keep using local models even when/if AI becomes available in China where i work).

I think Jon's strategy of sticking to Apple Intelligence is fine, as the various APIs and quickly changing fortunes of foundation model developers means there is still a lot of changes that come thick and fast , this is still not a stable landscape (though probably OpenAI and Google at least won't disappear overnight)...

[Jon]

Note that, at least in my tests, all of the intrinsic Apple Intelligence features run locally. Only Compose, which uses ChatGPT, requires an internet connection.

Jon
Sonny Software