An open-source utility called RemoveMacAI promises to disable Apple Intelligence features and remove downloaded models without disabling macOS System Integrity Protection. The headline figure is around 12GB, but that is the creator’s reported model footprint, not a verified saving every Mac owner can expect.
The RemoveMacAI GitHub repository is published under the account omlahore with an MIT licence. Its first public release was on 29 September 2026, and version 0.2.4 followed on 5 October. This is an existing project that has received attention and updates, rather than a tool launched today.
For people who do not use Apple Intelligence, the appeal is straightforward: more control over installed AI features and potentially less storage occupied by unused models. However, compatibility inconsistencies, delayed storage reclamation and the consequences for other applications deserve more attention than a single gigabyte figure.
Verification note: DIY AI reviewed the project’s documentation, release history and selected source files. We have not run RemoveMacAI on a Mac, independently measured storage savings or completed a security audit.
What RemoveMacAI actually changes
RemoveMacAI is not simply a command that deletes an “AI” folder. Its approach combines a configuration profile, restrictions on selected features and removal requests through Apple’s model asset service.
The profile requires approval in System Settings. In the source we inspected, it applies system-wide settings and redirects downloads for targeted model asset types to a local address designed to prevent those downloads from succeeding. That is a specific model-download control, not a firewall blocking every AI service.
The project says this approach leaves System Integrity Protection enabled and avoids directly modifying /System. Those distinctions matter, especially compared with instructions that tell users to weaken macOS protections before manually deleting protected files. They do not, by themselves, establish that the utility is risk-free.
The feature catalogue covers Siri, Writing Tools, Genmoji, Image Playground, Apple’s external AI integrations and several summarisation features. Its model targets also include Photos Clean Up, Spatial Photos and Xcode code completion.
Several features share underlying models, so the consequences extend beyond a single visible toggle. The project also warns about effects on applications using Apple’s Foundation Models framework. Developers should therefore check their own software’s on-device model dependencies before treating these assets as disposable.
Why “12GB removed” is not the same as 12GB reclaimed
The creator described approximately 12GB of models on their own Mac. That is useful context, but it is not a controlled before-and-after storage measurement.
Apple’s current requirements illustrate why a universal saving would be misleading. Its documentation lists up to 14GB of storage for compatible Macs with an M3 or later chip and at least 12GB of unified memory, and up to 8GB for other compatible Macs. These are Apple’s published requirements, not estimates of what RemoveMacAI will remove from a particular machine.
There is also a timing issue. The project’s changelog explains that releasing models through Apple’s asset service doesn’t necessarily return disk space immediately. macOS handles subsequent file removal, and System Settings can continue counting that storage for a while.
Recent fixes reinforce the need to check the outcome. Version 0.2.3 addressed a detection problem that could report zero installed models and consequently remove nothing. Version 0.2.4 changed how unexpected model sets are handled, allowing the utility to skip an incompatible set rather than abandon the entire removal attempt.
The meaningful result is an increase in usable free space, not simply a success message. A proper test should separately record feature state, the model inventory and available disk space, including any model sets the utility skipped.
The macOS compatibility information needs clarification
The project’s published support information has an important inconsistency.
The README documents support for Apple-silicon Macs running macOS 27 and explicitly describes macOS 26 and earlier as unsupported. However, the source code’s operating-system check accepts macOS 26, while the version 0.2.4 release notes describe a fix involving macOS 26.5.
An executable accepting an operating-system version is not the same as a clear support commitment. Until those statements are reconciled, macOS 26 users should not interpret the release note as blanket confirmation that their system is supported.
Anyone evaluating the utility should record the exact macOS build, Mac hardware and RemoveMacAI version. Results from one combination should not automatically be applied to another.
Inspect the planned changes before applying them
A utility that changes system-wide settings deserves more scrutiny than an ordinary application download. Review the release and its source before installation, and use a backed-up test Mac for an initial evaluation rather than a production machine.
The installer source we inspected downloads a release archive and a SHA-256 checksum, compares them and stops if they do not match. That can detect a mismatched download. It does not independently prove that the software is benign, because both files come from the same release source.
The project also documents build attestations. Build provenance and checksums are useful evidence, but neither substitutes for a security review or testing the effects on your own system.
After installing a release you have reviewed, these commands expose the version, current status, feature catalogue and proposed removal plan:
removemacai --version
removemacai status
removemacai features
removemacai off --dry-runThe dry-run path can write a temporary configuration profile for inspection, but it does not install that profile or delete the models. Read its output for anything the utility intends to leave behind.
For an actual evaluation, record free space before changing anything, check again after macOS has completed its cleanup, and test the features and applications you rely on. Restoration should be part of the test, not an assumption made after something stops working.
Undoing the change is more than uninstalling the command
The revert implementation opens System Settings and asks the user to remove the RemoveMacAI profile. The project says the user’s own settings then apply again, and macOS downloads models when the relevant features are re-enabled.
That is not an instant, offline restoration of the deleted assets. Allow for downloads and verify that dependent features work again. Removing the executable alone should not be confused with removing its installed configuration profile.
Storage control is not a complete privacy guarantee
Downloaded model files are not the same thing as personal conversations, documents or a complete record of AI activity. Removing model weights does not establish that personal records have been erased or that no AI-related data can leave the Mac.
Apple describes Apple Intelligence as combining on-device processing with Private Cloud Compute for some requests. Third-party AI applications and browser-based services are separate again. Disabling Apple’s built-in features is therefore not a universal “no AI” policy.
As our guide to what a local LLM keeps on-device explains, model storage, local processing and external connectivity are different questions.
Apple’s Apple Intelligence Report, available through Privacy & Security settings, provides a separate way to inspect requests to Private Cloud Compute. It should not be mistaken for a record of every AI service used on the computer.
What consumers and IT teams should do first?
For users who mainly want fewer AI features, Apple documents individual controls in macOS 27, including turning off Siri and disabling summaries in Mail, Messages and Notifications. Start with the settings for the feature causing the problem.
Those controls address behaviour. They should not be interpreted as a promise that all associated model storage will immediately disappear.
For organisations, Apple’s device-management documentation provides restrictions for selected intelligence features and external integrations. Availability depends on the operating-system version, management service and applicable device requirements.
Administrators should separate three objectives: disabling unwanted features, controlling external AI integrations and reclaiming storage. A policy that achieves one does not automatically achieve the others. Test any additional removal profile alongside existing management policies before considering wider deployment.
DIY AI’s view
RemoveMacAI highlights a legitimate user-control question: how easily should Mac owners be able to remove large AI components they do not use?
The project offers an inspectable approach, but its compatibility mismatch and recent removal fixes argue against presenting it as a guaranteed 12GB cleanup. For technically confident users, it is worth evaluating on a test machine. For everyday users and managed fleets, supported settings and carefully scoped policies remain the more cautious starting point.
The result that matters is not the largest storage number in a headline. It is verified space reclaimed, understood feature losses and a restoration process that works on the exact Mac being used.
