Hello everyone. I am very excired to announce AFM Studio, the best way to test AI models and providers on macOS and iOS. The project is open source and contributions are welcome and encouraged. You can learn more and try it out by visiting the repo at github.com/Techopolis/AFM… We have support for the system model, PCC, and Core AI models currently. This even includes our own model registry that will be updated often. Enjoy! #buildInPublic, #OpenSource, #AppleAI

So I tested "normal" English text generation from an open-source open-weights model with no weird incentives, and it performs really well! Just needs slight fine-tuning on the prompt. Looks like the big labs are in fact nerfing English for "fingerprinting" or whatever! Open source for the win! x.com/NatashaTheRobo…

One of my other free apps, Leveler, just got updated to 1.3 on the App Store and now includes Breath Control and Mouth De-Click!

I converted @liquidai’s LFM3-VL-2B to Apple’s Core AI and shared it on @huggingface. Ask about a picture, get an answer on your Mac. Swift package included.

What do you think? I’d love to hear what you’d build with it.

huggingface.co/tayarndt/LFM3-…

#OnDeviceAI #Swift

Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.

They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.

I converted Microsoft’s Mage-VL-Cowork for Apple’s Core AI, and I’ve now published my experimental version on Hugging Face!

Mage is a model that looks at screenshots and proposes computer actions. My part was converting it and testing how closely the results matched the original. Some tests produced matching actions. Others showed differences or missed the target, so there’s still work to do.

I’ve shared the converted model assets, a small test you can run with prepared inputs, and the results so other developers can explore the work too. It’s early, and it won’t control your computer out of the box.

Let me know what you think! What other models would you like to see converted to Core AI?

huggingface.co/tayarndt/Mage-…

#CoreAI #OnDeviceAI

Beyond The Gallery 2.4 is now available with iPhone Duo support. Check out our latest release and share your favorite Shortcuts.

Releasing Muse Ember today -- our smallest model that delivers higher intelligence per dollar than others in its class. Built for token-heavy work that agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing. Available today on the Meta Model API and in Muse Code.

My thoughts on how Language Models are good at everything except Language 🤔 natashatherobot.com/p/language-mod…

For the first time in 27.3, Siri AI routes ~2% of English requests to a local model, instead of PCC, when online. Mostly calls and music.

Skimmer still runs first and does most on-device work. When your phrasing matches a regex, it handles the request with preset actions.

I plan to upload so many more models up to Hugging Face. Please let me know if you have trouble with them and I will work to make improvements.

Here is a Core AI build of Gemma 4 12B Text only at FP16. huggingface.co/mikedoise/gemm… Working on the full model next and will then quantize it.

Hey @stephenrobles. I just watched your video where you talk about running on-device AI through LM Studio. Have you looked at Muse Glimmer for your larger model instead of Llama? Both are from Meta, and Muse Glimmer is their newer model. Also, there is a cool app called oMLX which is a very nice server for running models. It will work with your same shortcuts. It has really good cacheing. I use it for all of my local inferrancing and should work great on your Mac Studio.

I personally believe that apps as we know them are dead. What's most important now is experiences. So, what experience does your app provide that is different from others. Just building an app now isn't enough. Does it provide an experience people want to continue to use and actually enjoy!

I've created Core AI versions of Google Gemma 4 E2B IT 4-bit and they are now on Hugging Face. There is a text only and text + Vision variant. These use Apple's official recipe, and they are 128k context. I still need to do more testing on them, but the collection can be found at huggingface.co/collections/mi…

Spotify is working on iPhone Duo support, as seen in the Spotify beta for iOS

So I've tried another proactive AI agent last night called Instinct. It is cool, but still limited. It has a few cool connectors, but you aren't able to add your own yet. I will be watching this one, but my favorites are still ChatGPT Dots, Grok Bot and Muse.

The RSS Reader Unread has iPhone Duo support.

VoiceOver and Zoom on the MacBook Pro with PROLED is amazing, but it has a few bugs. Zoom specifically lets you scroll with one finger but you can't scroll in lists with one finger while magnified yet. The dynamic island will also not bring down Spotlight either when zoomed in as well. VoiceOver I think, has some major accessibility wins. You do NOT need to interact in many places on the touch screen, and you can even jump keyboard focus to the touch cursor. Those are just a few things that are there. VoiceOver also does not see an empty dynamic Island either.

[App Directory] koedesk dlvr.​it/TWF45v

SoFi now has iPhone Duo support.