Ollama Filled My Disk: How I Moved the Models Directory Safely

Ollama Filled My Disk: How I Moved the Models Directory Safely

Eleven models later my 512 GB system drive had 9 GB free and Windows started complaining. Ollama does not warn you: it writes everything to ~/.ollama/models on the boot volume. Moving it takes five minutes and zero re-downloads.

Step 1 — See what you are actually storing

Related: Midjourney vs DALL-E vs Flux: A Practical Image-Tool Comparison Framework →

Related: Ollama Answers Get Cut Off Mid-Sentence: num_ctx vs num_predict Explained →

ollama list
du -sh ~/.ollama/models
ollama rm llama2:13b     # models I had not touched in months

I recovered 46 GB before moving anything. Old experiments are the cheapest win.

Step 2 — Linux / WSL2

Related: 9 Free AI Coding Tools Every Developer Should Try in 2026 →

Related: Ollama "connection refused on 127.0.0.1:11434": The 5 Causes I Have Actually Hit →

sudo systemctl stop ollama
sudo mkdir -p /mnt/data/ollama-models
sudo rsync -a --info=progress2 ~/.ollama/models/ /mnt/data/ollama-models/
sudo chown -R ollama:ollama /mnt/data/ollama-models

sudo systemctl edit ollama

[Service]

Environment="OLLAMA_MODELS=/mnt/data/ollama-models"

sudo systemctl daemon-reload && sudo systemctl start ollama ollama list # verify BEFORE deleting the original rm -rf ~/.ollama/models

The chown line matters: the daemon runs as the ollama user and silently re-downloads everything if it cannot read the directory. That happened to me on the first attempt.

Step 3 — Windows

Related: Sora 2 Review: OpenAI's Video Model Is Finally Useful for Real Work →

Related: Ollama Error "model requires more system memory": How I Fixed It in 10 Minutes →

Move the folder, then set a user environment variable:

robocopy "$env:USERPROFILE\.ollama\models" "D:\ollama-models" /E /MOVE
[Environment]::SetEnvironmentVariable("OLLAMA_MODELS","D:\ollama-models","User")

Quit Ollama from the tray icon and reopen — restarting the app is not optional, the variable is read at startup.

Step 4 — macOS

Related: How to Clone Your Voice with AI in 2026 (Free and Paid Options) →

Related: Ollama Running Slow? 7 Fixes That Actually Worked on My Machine →

launchctl setenv OLLAMA_MODELS "/Volumes/External/ollama-models"

External SSD caveat from my Mac mini: if the drive is not mounted at login, Ollama creates an empty directory and shows zero models. Internal storage or an always-connected Thunderbolt SSD only.

Step 5 — Verify before you delete

Related: DeepSeek-R1 Shows Its <think> Tags in the Output — Here Is How I Strip Them →

ollama list
ollama run llama3.1:8b "say ok"
df -h

Only then remove the original directory.

My layout

Rule I now follow: keep at most six models, one per job. Storage discipline beats a bigger disk.

FAQ

Will this break existing models? No. Blobs are content-addressed; a plain copy is enough as long as permissions are right.

Can I put models on a network share? Technically yes, in practice load times were awful over gigabit. Local SSD only.

A download died from lack of space — what now? Clear the partial blob first, as described in fixing Ollama pull errors. Model sizing guidance is in the local LLM guide and the ultimate AI tools guide.