Where apps hide your files
Ollama saves every model you pull into a hidden folder in your home directory, not in Applications or Documents. You can open it, see which models take the space, and move the whole thing to an external drive.
~/.ollama/models
That is the default on macOS, per Ollama's own FAQ. The folder starts with a dot, so Finder hides it. To open it, choose Go > Go to Folder (or press ⇧⌘G), paste ~/.ollama/models and press Return. From Terminal:
open ~/.ollama/models
If the folder is not there, either you have not pulled a model yet, or OLLAMA_MODELS points somewhere else (see the section on moving it below). To check whether the variable is set for apps launched from the Dock:
launchctl getenv OLLAMA_MODELS
An empty reply means it is not set that way, and the default applies.
Ollama stores models in a layout borrowed from container registries, so you will not see friendly file names like llama3.gguf.
Path under models/ | What it holds |
|---|---|
blobs/ | The actual data, in files named sha256- followed by a long hash. The biggest ones are the model weights. |
manifests/ | Small files without an extension, one per model tag, listing which blobs make up that model. |
manifests/registry.ollama.ai/library/<model>/<tag> | Where a model from the official library lands: a folder named after the model, holding a file named after the tag. |
The exact nesting inside manifests/ is described by community write-ups rather than by Ollama's documentation, so treat it as the current layout and not a promise. Models that share layers reuse the same blob, which is why deleting files by hand is a bad idea: one blob can belong to several tags.
The rest of ~/.ollama is small. Ollama's macOS page says the folder holds models and configuration, and that ~/.ollama/logs keeps server.log (the most recent server logs) and app.log (the desktop app's). The FAQ also puts your Ollama public key at ~/.ollama/id_ed25519.pub. Leave both id_ed25519 files alone and never share the one without .pub: it is the private half of the key pair Ollama uses to identify you to ollama.com.
The desktop app also has a folder at ~/Library/Application Support/Ollama, which Ollama's uninstall steps list separately from ~/.ollama. It holds the app's own data, not your models. On a Mac checked for this guide it contained only browser-style caches and a tiny config.json. If you went looking for gigabytes there and found almost nothing, that is why.
A single model can be anything from under a gigabyte to tens of gigabytes. List what you have, with sizes, using the supported command:
ollama list
To see how much the folder uses in total:
du -sh ~/.ollama/models
Use Ollama, not Finder:
ollama rm <model-name>
It removes the model's manifest, and blobs that no other model uses are cleaned up with it. Deleting individual sha256- files in Finder can leave a manifest pointing at a missing blob, and the next ollama run will then fail or download again. If you want to clear everything, quit Ollama from its menu bar icon first, then move ~/.ollama/models to the Trash in Finder. You will have to pull each model again.
Ollama reads the OLLAMA_MODELS environment variable. On a Mac, where Ollama runs as an app, the FAQ says to set it with launchctl and then restart the app:
~/.ollama/models into it in Finder.launchctl setenv OLLAMA_MODELS "/Volumes/FastDrive/ollama-models" (use your own path).ollama list to confirm your models appear.Three cautions. Setting the variable in ~/.zshrc only affects Ollama when you start it from that shell with ollama serve, not the menu bar app. launchctl setenv does not survive a restart of the Mac, so if the models seem to vanish after a reboot, run the command again or set the variable in a launch agent. And recent versions of the app have their own Settings window; if yours offers a model location there, that is simpler, but the FAQ documents only the environment variable.
If you installed Ollama with Homebrew and run it as a background service, the service has its own environment, so the variable has to be set where that service is defined. brew services info ollama shows how yours is launched.
Copy blobs and manifests together, keeping the folder structure. One without the other is useless. Time Machine includes hidden folders, so it backs this one up unless you exclude it, and excluding it is reasonable because every model can be downloaded again. If bandwidth matters more than disk, copy the models folder to the new Mac's ~/.ollama/ and run ollama list there to check.
~/.ollama is a hidden dot-folder, and Spotlight does not return results from hidden folders. The blobs also have hash names with no extension, so even a perfect index would not match "llama" to them.
The friendly names live in the manifest folders, so that is what to search for. Typing a model name such as llama3.2 into Everywhere finds its manifest folder under registry.ollama.ai, and typing sha256- lists the blob files, even though both sit in a hidden folder. Press ⌘↵ to reveal one in Finder. It matches file names only, so it cannot tell you which blob belongs to which model; ollama list and the manifests do that.
In ~/.ollama/models, split into blobs and manifests. The folder is hidden, so open it with Go to Folder (⇧⌘G) or open ~/.ollama/models.
Quit Ollama, move the models to the new location, run launchctl setenv OLLAMA_MODELS "<new path>", then start the app again. Setting it only in your shell profile does not affect the menu bar app.
Prefer ollama rm <model>, which removes the manifest and unshared blobs together. Deleting blobs by hand can break other models that share them.
No. That folder holds the app's own cache and settings. Models live in ~/.ollama/models unless OLLAMA_MODELS points elsewhere.
Everywhere keeps track of every file on your Mac — including the folders Spotlight hides and drives you have unplugged. Free for a day, then $19 once.
Version 2.20.3 · macOS 13+ · 4.7 MB · notarised