How to Move Ollama Models to Another Drive (macOS, Linux, Windows)

Bottom Line

• One variable does it: OLLAMA_MODELS. Point it at the new directory and restart Ollama.
• Move the files yourself. Setting the variable does not migrate anything. Move both blobs/ and manifests/, or your library disappears.
• Set it where the service reads it. A shell export never reaches a systemd daemon or a macOS app. This is the single most common failure.
• macOS gotcha the docs skip: launchctl setenv is lost on reboot. Use a LaunchAgent if you want it to stick.
• Linux gotcha: ownership. The standard installer runs the daemon as the ollama user, which needs chown -R ollama:ollama on the target.
• Defaults: ~/.ollama/models on macOS, /usr/share/ollama/.ollama/models on Linux, C:\Users\%username%\.ollama\models on Windows.
• External drives work, with a caveat. If the drive is not mounted at service start, Ollama behaves as though your models are gone.

What You Are Actually Moving

Open the models directory and you will find exactly two things:

models/
├── blobs/       # the weight data — this is the tens of gigabytes
└── manifests/   # which blobs make up each tagged model

Both must move together. The manifests are small and easy to forget. Move only the blobs and Ollama sees no models at all, even though the data is right there. Move only the manifests and every model reports missing layers.

This also explains why you should move rather than re-pull. The blobs are content-addressed, so moving them keeps every tag you already have without re-downloading anything.

macOS

The documented steps:

# 1. Create the target directory on the new drive
mkdir -p /Volumes/BigDrive/ollama-models

# 2. Move your existing library
mv ~/.ollama/models/blobs ~/.ollama/models/manifests /Volumes/BigDrive/ollama-models/

# 3. Set the variable
launchctl setenv OLLAMA_MODELS "/Volumes/BigDrive/ollama-models"

# 4. Restart the Ollama application

Now the part the FAQ does not tell you. launchctl setenv sets the variable for the current boot only. It is not persistent. Restart the Mac and the value is gone, Ollama falls back to ~/.ollama/models, and your library appears to have vanished. Nothing is lost — the daemon is simply reading the old path again.

To make it permanent, run the same command from a LaunchAgent at login. Create ~/Library/LaunchAgents/com.user.ollama-models.plist:

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN"
  "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
  <key>Label</key>
  <string>com.user.ollama-models</string>
  <key>ProgramArguments</key>
  <array>
    <string>/bin/launchctl</string>
    <string>setenv</string>
    <string>OLLAMA_MODELS</string>
    <string>/Volumes/BigDrive/ollama-models</string>
  </array>
  <key>RunAtLoad</key>
  <true/>
</dict>
</plist>

Then load it with launchctl load ~/Library/LaunchAgents/com.user.ollama-models.plist. The LaunchAgent approach is the standard replacement for the withdrawn /etc/launchd.conf, and it is not specific to Ollama.

Linux

The daemon runs under systemd, so the variable belongs in the unit.

# 1. Create the target and give the daemon ownership
sudo mkdir -p /mnt/bigdrive/ollama-models
sudo chown -R ollama:ollama /mnt/bigdrive/ollama-models

# 2. Move the existing library
sudo mv /usr/share/ollama/.ollama/models/blobs \
        /usr/share/ollama/.ollama/models/manifests \
        /mnt/bigdrive/ollama-models/
sudo chown -R ollama:ollama /mnt/bigdrive/ollama-models

# 3. Edit the unit
sudo systemctl edit ollama.service

Add this under [Service]:

[Service]
Environment="OLLAMA_MODELS=/mnt/bigdrive/ollama-models"

Then reload and restart:

sudo systemctl daemon-reload
sudo systemctl restart ollama

Two things break this reliably.

First, ownership. The standard installer creates a dedicated ollama user, and that user needs read and write access to the directory. sudo chown -R ollama:ollama <directory> is the documented fix. Run it again after you move files in, because mv across filesystems can carry the old ownership.

Second, traversal. The ollama user needs execute permission on every directory in the path, not just the last one. A mount point at /mnt/bigdrive owned by root with mode 700 will block the daemon even when the final directory is owned correctly. Check with sudo -u ollama ls /mnt/bigdrive/ollama-models — if that fails, the daemon will fail the same way.

Windows

1. Quit Ollama from the taskbar. Do not just close the window.
2. Open Settings and search for environment variables.
3. Click "Edit environment variables for your account".
4. Create OLLAMA_MODELS with the new path, for example D:\ollama-models
5. Click OK, then restart Ollama.

Move blobs and manifests out of C:\Users\%username%\.ollama\models into the new directory before you restart.

Quitting from the taskbar matters. Closing the window leaves the background service running with the old environment, and it will keep writing to the old path while you conclude the setting is broken.

Symlinks and Bind Mounts: When to Use Them Instead

The environment variable is the supported route. Two alternatives exist and each has a right moment.

Neither symlinks nor bind mounts are documented by Ollama. They work because they operate below the application, not because they are supported. If you file a bug report, mention that you are using one.

When It Does Not Work

Do You Need a Bigger Drive Instead?

Moving models buys you the capacity of the drive you moved to. If that drive is also small, you are solving the problem twice. Our model storage sizing guide has the arithmetic: 1TB is the floor, 2TB is the right default for a single-GPU box, and 4TB once you are past 48GB of VRAM. The multiplier that catches people out is not model count. It is keeping Q4, Q6 and Q8 of the same model while you decide which one to use.

See Also

• How Much Storage Do You Need for Local Models? — drive sizing, load-time arithmetic, and what to buy
• Ollama vs llama.cpp — llama.cpp takes a path argument and has no equivalent variable
• Using Hugging Face Models With Ollama — the other download cache that fills your boot drive, controlled by HF_HOME
• Ollama Crash Course for Local Agents — the rest of the operational basics
• Ollama vs LM Studio — how the other popular runtime handles model storage

Sources

• Ollama FAQ (docs.ollama.com) — default directories per operating system, OLLAMA_MODELS, the launchctl setenv and systemctl edit procedures, and the chown -R ollama:ollama permission requirement; read 2026-08-25
• The blobs/ and manifests/ layout was confirmed on a macOS install on 2026-08-25
• launchctl setenv is not persistent across reboots; the LaunchAgent pattern is the standard replacement for the /etc/launchd.conf support withdrawn in OS X 10.10
• Symlink and bind-mount behaviour is general operating-system behaviour, not documented Ollama support