Odysseus AI Cookbook Dependencies Crashed
Last updated: June 6, 2026
If Cookbook dependency downloads crash, or VRAM detection looks wrong, split the problem into shell tools, operating system route, GPU visibility, and the specific serve engine you are trying to install.
Quick answer
Do not reinstall the whole app first. Check tmux, Git Bash or WSL, GPU visibility, and the exact dependency that crashed. On Windows, use Ollama first if you just need a working local model.
1. Capture the Actual Failure
Start with logs instead of guessing. The failing layer matters more than the word crashed in the UI.
docker compose logs --tail 160 odysseus
docker compose psFor native installs, check the terminal where you started Odysseus and the Cookbook dependency panel output.
2. Check Required Shell Tools
Cookbook downloads and serves often rely on shell tools and tmux. Missing shell support is common on Windows and fresh Linux installs.
Linux / WSL
tmux -V
python3 --version
git --versionInstall missing tools before retrying Cookbook Dependencies.
Windows native
bash --version
py -3.12 --version
git --versionGit for Windows provides bash.exe, which the official README calls out for full Cookbook and shell-tool workflows.
3. Install Missing Linux Tools
Native Linux and WSL users should install baseline packages before trying model downloads again.
sudo apt update
sudo apt install -y git python3 python3-venv tmuxUse your distro equivalent if you are not on Ubuntu or Debian.
4. Windows: Separate App Startup from Local GPU Serving
The core app can run native on Windows, but local GPU serving with vLLM or SGLang is a Linux or WSL path. If Cookbook dependency installs fail on Windows, use a simpler local backend first.
Fastest working local model
Install Ollama natively on Windows and point Odysseus to http://localhost:11434/v1.
Linux-style local serving
Use WSL or Linux when the dependency path expects Linux GPU libraries or tmux.
Docker app plus host Ollama
Run Odysseus in Docker and connect to Windows Ollama with http://host.docker.internal:11434/v1.
5. Debug Wrong VRAM Detection
Cookbook reports what the current runtime can see. A 16 GB GPU can appear smaller if the app is seeing an iGPU, WSL abstraction, Docker limit, or a runtime without the expected GPU device.
NVIDIA
nvidia-smi
docker compose exec odysseus nvidia-smi -LCompare host visibility with container visibility.
AMD / ROCm
rocminfo
ls -l /dev/kfd /dev/dri/renderD*Check whether ROCm and render devices are visible to the runtime.
If Docker cannot see the GPU, use the GPU troubleshooting guide.
6. Retry One Dependency at a Time
After fixing shell tools or GPU visibility, retry only the dependency that failed. Do not switch model runtime, GPU overlay, and install method all at once.
- - Reopen Odysseus as an admin.
- - Go to Cookbook, then Dependencies.
- - Reinstall the failed runtime, such as llama.cpp or vLLM.
- - Serve a small model first.
- - Capture logs if it still crashes.
7. Use Ollama as a Practical Fallback
Cookbook serving is useful, but it does not have to block your first successful chat. Ollama is often the fastest local backend on Windows and macOS, and a stable fallback on Linux.
ollama pull llama3.1:8b
ollama listIf the endpoint fails, use the Ollama connection guide.
Verify Against Official Issue Context
This page tracks the pattern from official issue #758 and the README notes about tmux, Windows shell support, WSL, and local GPU serving. Check the official issue and README for the latest maintainer notes.
FAQ
Why does Cookbook Dependencies say crashed?
Common causes include missing tmux, missing Git Bash on Windows, WSL not being used for Linux-only model serving, dependency installation failure, or GPU and VRAM detection not matching the runtime you are using.
Does Cookbook need tmux?
Yes. The official README says Cookbook needs tmux for background model downloads and serves on native Linux. Install tmux before relying on Cookbook dependency workflows.
Why does VRAM detection show the wrong amount?
Cookbook can only detect what the current runtime exposes. Docker, WSL, host drivers, iGPU selection, AMD ROCm permissions, or NVIDIA passthrough can make the visible VRAM differ from the physical GPU.
Can Windows native Odysseus serve vLLM or SGLang locally?
The official README says local GPU serving of vLLM and SGLang needs Linux or WSL. On Windows, native Ollama is usually the easiest local model backend.
Should I reinstall everything when Cookbook crashes?
No. First identify whether the crash is dependency install, shell/tmux, GPU detection, or model runtime. Reinstall only the failing serve engine or missing dependency path.
Installation support
Need hands-on installation support?
Setup Helper is the self-service planner. Installation Support is for users who want a human to review the route, diagnose logs, or get Odysseus running. We confirm scope and price before any payment.
Related Guides
Fix CUDA runtime and CPU fallback after Docker GPU passthrough works.
Check host drivers, Docker passthrough, and model backend visibility.
Install tmux, CUDA or ROCm, and native Linux dependencies.
Use Docker, WSL, native Windows, or Ollama depending on the goal.
Use Ollama when Cookbook serving is blocking your first working chat.
Return to the main setup issue checklist.