How to Completely Uninstall Ollama on Mac Without Losing Custom Models

Before uninstalling: list models, running jobs, and the real model directory

Ollama on macOS can include the app, a command-line entry, a background server, logs, keys, and a model library. The default model directory is commonly under ~/.ollama/models, but OLLAMA_MODELS can move it to another disk. Deleting only the application can therefore leave most of the occupied space behind; deleting ~/.ollama blindly can remove far more than downloaded weights.

While Ollama still runs, use ollama ps to see active models and ollama ls to list stored names and sizes. Record the install method — the macOS app, Homebrew, or another package — and inspect the environment or service configuration for OLLAMA_MODELS. Measure both the default and configured location, but do not assume a large directory is disposable until model provenance is known.

1. Run ollama ps and stop work that still depends on a loaded model.
2. Run ollama ls and save the names, tags, and sizes you may want later.
3. Check whether OLLAMA_MODELS points to a custom location or external drive.
4. For every custom model, save the output of ollama show --modelfile model:tag and keep its source GGUF, adapter, prompt files, and licence information.
5. Record whether Ollama came from the official macOS app or Homebrew so removal follows the same owner.

A model name in ollama ls does not prove it can be downloaded again. Imported and locally created models may exist only on this Mac.

Choose model cleanup or full uninstall

If Ollama still serves a useful workflow, remove individual downloaded models with ollama rm rather than uninstalling the runtime. This preserves configuration and avoids losing track of custom paths. The model-storage guide covers that narrower job and is the right answer when disk pressure — not the software itself — is the problem.

Full uninstall is appropriate when you no longer need the runtime, are switching to another local model service, or want a deliberate reset. Decide separately whether the model library remains as an archive, moves to external storage, or is deleted after custom assets are verified.

Quit Ollama and remove it by the original installation method

Quit the Ollama macOS application from its menu and confirm no model generation remains active. For the official app, move Ollama.app from Applications to the Trash after it has stopped. If Homebrew installed Ollama, use Homebrew's uninstall command so its package records and linked files stay consistent. Do not combine both methods unless both installations actually exist.

Run command -v ollama before and after removal. The command-line entry may be a link installed by the app, a Homebrew binary, or another manually managed file. Inspect the resolved path and owner instead of deleting every executable named ollama. If another tool depends on an Ollama-compatible endpoint, update that tool explicitly rather than leaving it pointed at a vanished background service.

Decide what to do with downloaded and custom models

Downloaded public model weights are usually the largest category and can normally be fetched again. Their deletion cost is bandwidth and time. Custom models need a different standard: a Modelfile may refer to a local GGUF file, adapter, template, system prompt, or parent model, and the Ollama model store is not a substitute for preserving those source assets.

For a custom model, save the Modelfile, source model or adapter, build instructions, and any licence or provenance notes. Verify the saved definition can locate its source files. If a model was quantized or converted locally and the original process is not reproducible, treat its generated artifact as user work, not cache.

Only after that separation should you archive or remove the configured model directory. When OLLAMA_MODELS points to an external or shared location, deleting the default ~/.ollama/models may free little, while deleting the custom root may affect other users or services.

Review configuration, logs, and keys separately

The ~/.ollama directory can contain more than models. Logs are diagnostic and normally disposable after the process stops. Configuration and key files can represent identity or access used by Ollama workflows. Keep or back them up when moving to another Mac, switching runtimes, or preserving access to private resources.

Do not publish keys, Modelfiles with secrets, or private model paths in screenshots or support requests. Do not remove the parent ~/.ollama directory just to clear logs. Classify its children, remove the app-owned low-risk parts, and leave unknown or custom data for review.

AskClean follows this boundary by showing known model stores and models individually. It can recommend a public model for review based on redownloadability, but custom model directories and identity material must remain protected.

Verify the uninstall and reclaimed space

Check Activity Monitor or run pgrep -fl ollama to confirm the server is gone. Run command -v ollama and verify Applications no longer contains the app. Inspect the default model directory and any OLLAMA_MODELS location according to the keep/delete decision you recorded.

Test one saved Modelfile or source archive before deleting its original. Confirm applications that previously called localhost Ollama now fail clearly or have been reconfigured. Compare model-library sizes before and after; macOS System Data may take time to reflect the change.

If several model runtimes and custom directories are mixed together, AskClean's local scan can identify common Ollama, LM Studio, Hugging Face, PyTorch, and Whisper stores without reading model contents. It is most useful after you have decided which local AI workflows still matter.

Ollama uninstall safety map

Ollama changes quickly. Check current official macOS and CLI documentation, and verify the configured model root on the actual machine.

FAQ