All articles
Troubleshooting4 min·4 July 2026

Ollama won't load your models on Windows? Check your username for accents

You installed Ollama, pulled a model, and it fails to load with a cryptic error. If your Windows username contains é, è, ç or any accent, this is very likely why — and the fix takes two minutes.

If your name is Jérôme, Hélène, François or Cécile, there's a good chance your Windows session — and therefore your user folder — carries an accent: C:\Users\Jérôme. Most software doesn't care. Ollama, unfortunately, can: models are stored under your user profile by default, and accented characters in that path can prevent them from loading, even though the download itself succeeded.

The symptoms

  • ollama pull completes without error — the model appears to download fine.
  • ollama run (or any app using Ollama) fails to load the model, or hangs.
  • The error mentions the model path, a file that "cannot be found", or looks like garbled text where the accent should be.
  • The same setup works flawlessly on a colleague's PC whose username has no accent.

The cause

By default, Ollama stores its models in C:\Users\<you>\.ollama\models. Somewhere along the chain that loads a model from disk, the accented character in the path gets encoded one way and decoded another (UTF-8 vs. the legacy Windows code page). The result: the software looks for a file at a path that, from its point of view, doesn't exist. This class of bug is invisible in the US — usernames there rarely carry accents — which is why it survives so long.

The 2-minute fix: move the models to an accent-free path

You don't need to rename your Windows account (please don't — that breaks other things). Ollama reads an environment variable, OLLAMA_MODELS, that tells it where to store and load models. Point it at a simple ASCII path and the problem disappears.

  • 1. Create a folder at an accent-free location, e.g. C:\ollama-models.
  • 2. Open "Edit environment variables for your account" (search it in the Start menu).
  • 3. Add a new variable: name OLLAMA_MODELS, value C:\ollama-models.
  • 4. Quit Ollama completely (system tray icon → Quit) and relaunch it.
  • 5. Re-pull your model (ollama pull llama3.2 or from your app) — it now lands in the new folder and loads correctly.
Tip: if you had already downloaded models, you can move the contents of C:\Users\<you>\.ollama\models into the new folder instead of re-downloading — several gigabytes saved.

Why we know this bug so well

We build Locaible, a local AI assistant for French businesses, on top of local inference. Accented Windows usernames are the single most common setup issue we see in France — so our installer detects it and applies this fix automatically. If you'd rather have local AI that just works, without the command line, that's exactly what we made Locaible for.

Sovereign by architecture

Run AI without a single file leaving your office.

100% on-device, GDPR & EU AI Act compliant by design. See it on your own use case.