ollama local model failed to load.

[x5c20]

Can you help me? I tried setting it up according to the method provided by @Pseudosance. I am sure I followed the exact same steps, and I can access the ollama I started in WSL using the dcoker command, but I still cannot connect properly and it tells me that the model update failed.

[x5c20]

By checking the logs, I found the cause of the error.
Issue: EBUSY: resource busy or locked when renaming model-registry.json
Error:
Error: EBUSY: resource busy or locked, rename '/srv/data/model-registry.json.tmp' -> '/srv/data/model-registry.json'
at Object.renameSync (node:fs:1012:11)
Root Cause
This error occurs because the project uses a bind mount for the model registry file:
- ./backend/model-registry.json:/srv/data/model-registry.json
When running in WSL (Windows Subsystem for Linux) + Docker, the combination of:

• Windows NTFS file system
• WSL2 virtualized filesystem
• Node.js fs.renameSync() (atomic rename from .tmp to .json)

frequently causes EBUSY (resource busy or locked) errors. This is a well-known compatibility issue in WSL environments when performing atomic file operations on bind-mounted files.

Solution (Tested & Working)

Step 1: Clean up

./run.sh stop
docker compose down -v --remove-orphans
rm -f ./backend/model-registry.json ./backend/model-registry.json.tmp

Step 2: Edit docker-compose.yml (Main Fix)
Open docker-compose.yml and modify the backend service volumes section:

services:
  backend:
    ...
    volumes:
      - ./backend:/usr/src/backend
      # Comment out or remove this line:
      # - ./backend/model-registry.json:/srv/data/model-registry.json

Step 3: Restart

./run.sh start -q

Step 4: Verify

./run.sh status

Step 5: Configure Ollama

Go to Settings → AI Models in the web UI and set:

Provider: ollama
Base URL: http://host-gateway:11434/v1
Model: your pulled model (e.g. llama3.1)