Before the LLM era writing a grammar checker was complicated and time-consuming. One had to implement a natural language parser, a formal grammar for the language and special rules to identify mistakes and suggest corrections.
With modern LLMs, a text can be checked with a simple prompt. Here's an example of what mistakes AI discovers and how it corrects the text:
Not too bad. But if you don't want to send your data to the cloud and use a local model, the result is worse, according to my experiments. It doesn't catch all mistakes and completely messes up markup annotations. A solution to this problem might be the cloud-based model presented by Apple at WWDC26 (see Apple Foundation Model on Private Cloud Compute). I've tried it out and it is very good at (not only) grammar checking. Although it sends the text to the cloud, Apple's strong focus on privacy ensures your data remain fully private.
I turned the code from my experiments into an iOS app which is available on the App Store. The current version uses the local on-device model that comes with iOS 26. I'll update the app in September for iOS 27 (when the cloud model becomes publicly available). If you want to try out the cloud-based model now and are willing to install the macOS 27 beta, get in touch and I'll give you a beta version of the updated app.
Top comments (0)