Pared - remove unwanted Apple Intelligence models without disabling SIP
Summary
Pared is an open-source macOS tool (app, terminal wizard, and Nix modules) that lets users selectively disable Apple Intelligence features and remove unwanted on-device models without turning off System Integrity Protection.
View Cached Full Text
Cached at: 10/06/26, 07:27 PM
4evy/pared
Source: https://github.com/4evy/pared
Unzip the download and drag Pared to Applications.
Prefer the terminal? Run this to open the setup wizard:
curl -fsSL --proto '=https' --proto-redir '=https' \
https://github.com/4evy/pared/releases/latest/download/install.sh | sh
Requires Apple silicon and macOS 27 or newer. No Xcode or Homebrew needed.
Website · Documentation · More install options
Apple Intelligence, on your terms
Choose which Apple Intelligence features stay on your Mac, then remove models you no longer need. Pared includes a native app, terminal wizard, and Nix modules. System Integrity Protection stays enabled.
Keep what you use
Keep Writing Tools, turn off Genmoji, or switch everything off. Pared keeps shared models needed by features in its catalog that you leave on or unmanaged.
A few choices, then a review
- Choose your features. Set each one to On, Off, or App Default. A new policy starts with everything off; review your choices before saving
- Finish setup. Install Pared’s configuration profile in System Settings. It applies supported controls and blocks unwanted model downloads
- Review model removal. Pared shows which models can go before you confirm. Saving feature settings and removing models are separate actions
Want a feature back? Turn it on, install the updated profile, then request its models in Pared. For features Pared can’t download directly, use their Apple app.
Use
Open Pared or run pared wizard. Opening either changes nothing; feature
choices stay in a draft until you save.
See the app manual or command-line guide for setup and commands.
Configure it with Nix
Use the nix-darwin or Home Manager module to manage your choices. Both remove
eligible models during activation by default; set
programs.pared.cleanupOnActivation = false; to opt out. Install the generated
profile in System Settings to block future downloads.
See the Nix setup guide and option reference.
Build from source
With Xcode 27 or newer, run from this checkout:
tools/build-app.sh
open .build/Pared.app
For CLI and offline installation, see the command-line guide.
How it works
Pared uses Apple’s asset service to remove models. Read how profiles and model removal work for the implementation and compatibility limits.
Similar Articles
Remove and Disable Apple Macos27 AI Models Tool
RemoveMacAI is an open-source CLI tool that disables Apple Intelligence features on macOS 27 and removes the on-disk AI models, preventing them from being re-downloaded. It works via a configuration profile and Apple's asset service, with fully reversible changes and no network tracking.
Command-line tool quickly removes Apple Intelligence from macOS 27
RemoveMacAI is an open-source command-line tool that lets users selectively remove Apple Intelligence features from macOS 27 using approved configuration profiles, without disabling System Integrity Protection. It has gained 1,700 GitHub stars and shows user demand for greater control over Apple's AI features.
An open-source tool lets you delete 12GB of Apple Intelligence data on macOS
An open-source command-line tool called RemoveMacAI lets macOS users disable Apple Intelligence features, delete roughly 12GB of AI models from disk, and prevent them from being re-downloaded, since Apple removed the single settings toggle that once handled this.
Turn off and restrict access to Apple Intelligence features on Mac
This Apple support article provides instructions on how to turn off or restrict access to various Apple Intelligence features on macOS, including Siri AI, message summaries, and more, with options for parental controls.
@jakevin7: Model IQ Detector. Lately, all the big players keep dumbing down, making this a real must-have.
A macOS plugin that detects if AI models are being secretly downgraded or swapped in Codex by analyzing session fingerprints and providing verdicts on model integrity.