Inferencer for macOS v1.0 now available

Version 1.0 is now available for download on the Mac App Store. It’s been built with native code and in my testing it’s the fastest inferencer available on Mac, even with the added support for token inspection.

One of the good things about it being released on the official Mac App Store is that it fully supports sandboxing requirements, which means it only has access to folders you specifically allow it access to. Some models like Kimi K2 require the use of custom python tokenizers - which could, of course, be a vector for attack, so forcing sandboxing compliance is a good thing.

Regarding next features, we’ve already been asked to implement support for markdown rendering and configuring font sizes - which will be in the v1.1 update as soon as it’s approved by the App Store.

I’ll also be adding a public feature roadmap where you’ll be able to suggest and vote for improvements. If you have any suggestions in the mean time, I’d love to hear your thoughts.

Finally, as a reminder, all AI processing is done offline, directly on your device. No data is sent to online servers for maximum privacy.