Bundle Ollama models into single portable archives for transfer between machines.

Why would I need ollamazip?

pip install ollamazip

# With zstd compression support (recommended):
pip install ollamazip[zstd]

# With the graphical interface:
pip install ollamazip[gui]

# GUI in a native desktop window (via pywebview):
pip install ollamazip[gui-native]

# Pack a model into a portable .ollamazip file
ollamazip pack llama3:8b
ollamazip pack mymodel:latest -o mymodel.ollamazip
ollamazip pack qwen3:27b --compress gzip

# Unpack on another machine
ollamazip unpack llama3-8b.ollamazip
ollamazip unpack mymodel.ollamazip --name newname:v2
ollamazip unpack mymodel.ollamazip --verify

# Inspect an archive without unpacking
ollamazip list mymodel.ollamazip

# List installed Ollama models
ollamazip models

# Launch the graphical interface (opens in browser)
ollamazip gui
ollamazip gui --native    # open in a native window (requires gui-native extra)

ollamazip gui starts a small local web app (NiceGUI) that lets you:

• Browse installed Ollama models with sizes
• Pack any local model into a .ollamazip archive
• Rename or delete local models (with orphaned-blob pruning)
• Browse a folder of .ollamazip archives, inspect metadata, unpack, move, or delete them

By default it opens http://127.0.0.1:8734 in your browser. Pass --native to run in a standalone desktop window instead.

Local models	Archives
Browse, pack, rename, delete	Inspect, unpack, move, delete

• Single-file bundles: Packs the manifest and all content-addressed blobs into one archive
• Smart compression: Auto-selects zstd (if available) or gzip; override with --compress
• Blob deduplication: Skips blobs already present on the target machine during unpack
• SHA256 verification: --verify validates every blob against manifest digests
• Model renaming: --name newmodel:newtag imports under a different name
• Cross-platform: Works on macOS, Linux, and Windows; respects OLLAMA_MODELS env var
• No dependencies: Core functionality uses only the Python standard library; zstd is optional

Ollama stores models as a JSON manifest referencing content-addressed blobs (SHA256). ollamazip pack reads the manifest, collects all referenced blobs, and bundles them into a tar archive. ollamazip unpack extracts them into the target machine's Ollama model store.

ollamazip discovers Ollama's model directory automatically. The candidates, probed in order, are:

1. $OLLAMA_MODELS if set (hard override; ~ and $VAR are expanded).
2. The per-user store: ~/.ollama/models on macOS and Linux, %USERPROFILE%\.ollama\models on Windows.
3. Linux only: /usr/share/ollama/.ollama/models — the default for the official Ollama systemd-service install, where Ollama runs as a dedicated ollama system user.

If candidate (2) is empty/missing on Linux but (3) contains models, ollamazip uses (3). Run ollamazip models to see which path was picked; if the list is empty, it prints every path that was searched.

If you installed Ollama via the official Linux script, the model store at /usr/share/ollama/.ollama/models is owned by the ollama system user. As a regular user you can read that directory (so ollamazip pack and ollamazip models work), but you cannot write to it. To install models with ollamazip unpack, pick one:

• Re-run with sudo: sudo ollamazip unpack <archive>
• Add yourself to the ollama group and grant group write:

  sudo usermod -aG ollama "$USER"   # log out and back in
  sudo chmod -R g+w /usr/share/ollama/.ollama/models

• Move Ollama's storage to a directory you own and tell the service:

  sudo systemctl edit ollama.service
  # add under [Service]:
  #     Environment="OLLAMA_MODELS=/path/you/own"
  sudo systemctl restart ollama
  export OLLAMA_MODELS=/path/you/own

ollamazip prints similar guidance whenever a write operation (unpack, delete, rename) hits a permission error on the system path.

Fresh-install note: if you run ollamazip unpack before installing Ollama as a systemd service (so neither candidate dir contains models yet), ollamazip writes to ~/.ollama/models. The systemd-service Ollama won't see those models — point the service at your user dir with the systemctl edit recipe above, or move the unpacked files into /usr/share/ollama/.ollama/models after installing the service.

Copyright © Horst Herb.

Licensed under the GNU Affero General Public License v3.0 or later. AGPL-3.0 means you are free to use, modify, and redistribute this software, but any modified network-facing deployment must make the corresponding source available to its users.