Home / Guides / Why Is System Data So Big on My Mac? Local AI Models Are the Usual Suspect
Why Is System Data So Big on My Mac? Local AI Models Are the Usual Suspect
Before you start
The commands below are provided as-is, without warranty of any kind. Deletion is permanent: rm does not use the Trash and there is no undo, so make a backup first (for example with Time Machine). Paths and command names vary by tool version and how you installed it. Use at your own risk; Yogi Software Works is not liable for data loss or other damage resulting from these commands. See our Terms.
If System Data is huge on your Mac and you have used Ollama, LM Studio or Hugging Face tools, local AI models may account for a large part of it. Models are multi-gigabyte downloads that tools store in hidden folders in your home folder, so they are easy to forget and do not appear in the app list. This guide shows what Apple says System Data is, how to measure the usual model folders from Terminal, and how to remove models you no longer need.
Disclosure: I'm the developer of AiKiller (Yogi Software Works), a paid Mac app that can do part of this cleanup. The manual steps below work without it.
What Apple says System Data is
Apple's Storage documentation describes System Data as files that do not fall into the other categories, primarily files and data used by the system, such as log files, caches, VM files and other runtime resources, plus temporary files, fonts, app support files and plug-ins. Apple also says you cannot manage the contents of this category from that screen, that macOS manages them, and that the size varies with the current state of your Mac.
Apple's page does not name hidden folders such as ~/.ollama, and we did not test how macOS labels model folders in the Storage bar. It may differ by macOS version. That is why this guide measures the folders directly instead of relying on the bar.
To see the bar itself, choose Apple menu → System Settings → General → Storage.
Step 1: Measure the usual local AI model folders
This read-only command prints the size of each folder that exists. Folders you do not have are skipped:
du -sh ~/.ollama ~/.cache/huggingface ~/.cache/lm-studio ~/.lmstudio ~/.cache/vllm 2>/dev/nullWhat each one is:
~/.ollama: Ollama. Its FAQ gives~/.ollama/modelsas the macOS model location and says theOLLAMA_MODELSenvironment variable can point it somewhere else. The same folder holds the Ollama public key fileid_ed25519.pub.~/.cache/huggingface: the default Hugging Face home. Downloaded repositories are in itshubfolder. The location can be changed withHF_HOMEorHF_HUB_CACHE(see the environment variables documentation). Libraries such astransformersandvllmuse this cache layout, per the Hub documentation.~/.lmstudioand~/.cache/lm-studio: LM Studio. Its documentation shows models under~/.lmstudio/modelsand the app lets you change the models directory. AiKiller also checks~/.cache/lm-studio/models, which may be used by older versions. See the LM Studio and Hugging Face guide.~/.cache/vllm: the vLLM cache location in AiKiller's detection rules. We did not find it in official documentation, so treat it as a location to look at, not a confirmed one.
If you moved a model folder with one of the settings above, measure the new location too.
Step 2: Find the biggest folders in your home folder, including hidden ones
Hidden folders (names that start with a dot) are included in this listing. It can take a while on a large home folder, and macOS may ask whether Terminal may read some folders. The command only reads sizes:
du -hd1 ~ 2>/dev/null | sort -hr | head -15Then look inside the biggest candidates. ~/.cache holds many tools' caches, and ~/Library holds app support files and caches:
du -hd1 ~/.cache 2>/dev/null | sort -hr | head
du -hd1 ~/Library 2>/dev/null | sort -hr | headStep 3: Check Time Machine local snapshots
System Data can also include local snapshots and other items that have nothing to do with AI tools. To list local snapshots on the startup volume (an empty list means none were found):
tmutil listlocalsnapshots /tmutil also has deletelocalsnapshots and thinlocalsnapshots subcommands. Read man tmutil before using them, because macOS manages these snapshots itself and this guide does not cover removing them.
Step 4: Remove models you no longer need
Deletion is permanent. Back up any custom models you created locally before you remove anything.
Ollama
List the models and remove them with Ollama's own commands (see the CLI reference):
ollama ls
ollama rm <model>AiKiller also uses ollama rm for Ollama models. How the model folder is organized, and why its files should not be deleted by hand, is covered in Where Ollama Stores Models on Mac, and How to Delete Them. For a full removal of the app and the ~/.ollama folder, see How to remove Ollama from Mac.
Hugging Face
With the hf command-line tool from the huggingface_hub CLI, list the cache and remove a repository. --dry-run previews what would be deleted:
hf cache ls
hf cache rm model/<org>/<name> --dry-runOlder versions of the tool used different command names, so run hf cache --help and compare with the cache documentation for your version. More detail, including what not to delete, is in the LM Studio and Hugging Face guide.
LM Studio
LM Studio keeps model files in its models folder; see the LM Studio and Hugging Face guide for how to list and delete them.
After removing models, run the Step 1 command again to confirm the folders shrank.
What this guide does not cover
System Data has other contents (caches, logs, snapshots, app support files and more), and we have not tried to account for all of them. We did not check how a given macOS version classifies model folders, and we did not measure any particular Mac. If the folders in Step 1 are small, the space is somewhere else, and Step 2 is the way to find it. Other local AI apps may use folders not listed here; check each app's documentation.
Do this in one scan with AiKiller (free)
AiKiller's Local Models screen lists the models it finds in the Ollama, Hugging Face and LM Studio locations above with their sizes, so you can see them in one place. Its Ollama list reads the manifests under ~/.ollama/models/manifests/registry.ollama.ai/library, so models stored under other registry paths may not appear, and it checks the default folders only. Scan Mode is read-only: it lists each item with its path and size. Nothing is deleted until you switch to Clean Mode, select the items you want and type a confirmation code. Deletion is permanent and does not use the Trash, so back up anything you need first. AiKiller is free. Safety Shield (Applications screen only, on by default): when you remove an application, it keeps the credential files AiKiller recognizes, but recognition is limited to AiKiller's built-in detection rules, so other secrets may not be recognized. It does not apply to local model removal, which deletes everything you select.