Where Ollama stores models, and how to change the location

Ollama keeps models in a folder called .ollama/models in your home directory, except on Linux, where the install script's service keeps them under /usr/share/ollama. To put them somewhere else, set the OLLAMA_MODELS environment variable to a new folder and restart Ollama.

The default locations

System	Folder
Windows	`C:\Users\<you>\.ollama\models`
macOS	`~/.ollama/models`
Linux, installed with the script	`/usr/share/ollama/.ollama/models`
Linux, running `ollama serve` yourself	`~/.ollama/models`
Docker	`/root/.ollama` inside the container, usually a named volume

The folder starting with a dot is hidden by default. On a Mac, press Cmd+Shift+. in Finder to show it. On Windows, turn on hidden items in File Explorer's View menu.

What's inside

Two folders:

blobs holds the actual data: weights, templates and settings, each in a file named after its checksum, like sha256-.... The weights are one big file per model.
manifests holds small files that say which blobs make up each model name and tag.

Don't rename or delete blobs by hand. Use ollama rm to remove a model and let Ollama clean up. See how to update and delete models.

How much space you need

Plan for more than you think. A starter set of Qwen3 8B and gpt-oss 20B is about 19 GB. Add Llama 3.3 70B and it's 62 GB. Every build on our builds page includes at least a 1 TB SSD for this reason.

Keep models on an SSD if you can. A model loads from disk into memory every time it starts, so on a slow hard drive a 20 GB model takes noticeably longer to load. Once loaded, the drive doesn't affect speed.

Change the location on Windows

Quit Ollama from its tray icon.
From the Start menu, open Edit environment variables for your account.
Add a variable OLLAMA_MODELS set to your new folder, for example D:\ollama\models.
Start Ollama again.

More Windows setup in run AI locally on Windows.

Change the location on a Mac

launchctl setenv OLLAMA_MODELS /Volumes/External/ollama-models

Quit Ollama from the menu bar and reopen it. This setting doesn't survive a restart, so you'll need to run it again after rebooting, or use a login script. If the folder is on an external drive, Ollama can't see your models while the drive is unplugged. See run AI locally on a Mac.

Change the location on Linux

The install script runs Ollama as a systemd service under its own ollama user. Make the folder and give that user ownership:

sudo mkdir -p /data/ollama/models && sudo chown -R ollama:ollama /data/ollama

Then add the variable to the service:

sudo systemctl edit ollama.service

In the editor, add these two lines, save, and close:

[Service]
Environment="OLLAMA_MODELS=/data/ollama/models"

Restart the service:

sudo systemctl restart ollama

If Ollama fails to start afterwards, it's almost always permissions. Check with journalctl -u ollama and make sure the ollama user can reach every folder in the path. More in run AI locally on Linux.

Bring your existing models along

Setting OLLAMA_MODELS points Ollama at an empty folder, so your models seem to vanish. They're still in the old place. To keep them, quit Ollama, then move the contents of the old models folder, both blobs and manifests, into the new one before you start Ollama again. Then ollama list should show everything.

If you'd rather start fresh, delete the old folder once the new one works and pull what you need.

Recent desktop apps

Newer versions of the Windows and Mac apps also have a settings window. If yours has a model location option there, it does the same job as OLLAMA_MODELS.

Where to go next

Running low on space? See how to update and delete models.
All the commands: the Ollama cheat sheet.
Which models are worth the disk space on your card? Check every GPU.