Odysseus Troubleshooting
Top fixes from community reports — Docker GPU, missing models, ports, and Ollama.
Quick answer
Odysseus is early software (released May 2026). Most install issues have known fixes. Try the matching section below before opening a GitHub issue.
Docker can't see NVIDIA GPU
Symptom: Cookbook shows wrong GPU or 0 VRAM. Chat says no models.
terminal
Add NVIDIA device reservations to docker-compose.yml:
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: all
capabilities: [gpu]
Then: docker compose down && docker compose up -d --build
Also install NVIDIA Container Toolkit on Linux.Cookbook downloaded model but chat shows 'No models'
Symptom: Model downloaded to disk but not served or not visible in chat.
terminal
1. In Cookbook, click "Serve" on the downloaded model — wait for it to finish
2. Check terminal/logs for errors (HF token, VRAM insufficient)
3. Alternatively: install Ollama, pull a model, add endpoint in Settings
4. For Docker + Ollama on host: use http://host.docker.internal:11434/v1Port 7000 already in use
Symptom: Server fails to start — address already in use.
terminal
# Docker: edit .env
APP_PORT=7001
docker compose down
docker compose up -d --build
# Native: change port in uvicorn command
python -m uvicorn app:app --host 127.0.0.1 --port 8000Ollama connection fails from Docker
Symptom: Odysseus in Docker can't reach Ollama on localhost.
terminal
# Wrong (inside container):
http://localhost:11434/v1
# Correct:
Windows/Mac Docker: http://host.docker.internal:11434/v1
Linux Docker: http://172.17.0.1:11434/v1
Verify Ollama is running: ollama listPowerShell blocks launch-windows.ps1
Symptom: Execution policy error on Windows.
terminal
Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass
powershell -ExecutionPolicy Bypass -File .\launch-windows.ps1FAQ