Troubleshooting

The system check

Settings → System check tests each part of wanna.sh on your computer. Each line is Ready, Needs attention or Not working:

Check	If it isn’t ready
Graphics cards	No NVIDIA graphics card found: everything runs on the processor. Check the card is installed and its driver works (nvidia-smi in a terminal should list it). On WSL2, install the driver on Windows.
Chat engine	Not set up yet is normal: llama.cpp downloads itself the first time you chat. If it failed, check your internet connection and try again.
Pictures, video, voices and music	Not installed (chat-only install): run the installer again without --chat-only. Not included in the chat-only image: use ghcr.io/wanna-sh/wanna:latest instead of latest-chat. Can’t see a graphics card: update your NVIDIA driver, then restart wanna.sh.
Code sandbox	Install or start Docker. Being prepared means the sandbox image is building, which happens once. Code is off because wanna.sh runs in Docker: see Turning on Code.
Free disk space	Delete models you don’t use in My models, or move your data folder with WANNA_HOME.
System memory	Less than 16 GB works, but stick to models rated Runs great here.
Hugging Face	Can’t reach huggingface.co: installed models still work, but finding and downloading new ones needs internet.
Web search for Code	Needs Docker for the private search engine. If the last search failed, the error is shown.

Common problems

“Windows protected your PC” when installing

The Windows installer isn’t code-signed yet. Click More info, then Run anyway. See Install.

The Windows app says “wanna.sh couldn’t start”

The first start needs an internet connection to download Python and wanna.sh’s core parts. Press Try again. If it keeps failing, press Open the log folder and include setup.log and server.log when you report the problem. The logs are in %LOCALAPPDATA%\sh.wanna.desktop\logs.

The page says “The web UI hasn’t been built yet”

The interface wasn’t built. Run the installer again, or by hand: cd ~/wanna.sh/frontend && npm ci && npm run build, then restart wanna.sh.

wanna: command not found

~/.local/bin isn’t on your PATH. Add export PATH="$HOME/.local/bin:$PATH" to ~/.bashrc (or ~/.zshrc) and open a new terminal, or run ~/wanna.sh/backend/.venv/bin/wanna directly.

“Requests from web pages on … aren’t allowed”

A web page on another site tried to use wanna.sh, and wanna.sh refused, which is what protects you from websites you visit. If it’s your own web app, add its origin to WANNA_CORS. See Security.

“wanna.sh doesn’t answer to the name …”

You opened wanna.sh with a host name it doesn’t know, such as mypc.local or a domain behind a reverse proxy. Add the name to WANNA_ALLOWED_HOSTS, or use localhost or the computer’s IP address.

Port 7860 is already in use

Another program (or another copy of wanna.sh) is using it. Stop it, or start wanna.sh on another port: wanna --port 7861.

A model won’t start

• Look at its rating in Get models. If it’s Too big for this computer, choose a smaller version.
• Stop other models in My models (Stop all and free memory) and try again.
• For chat models, choose less Memory for conversation under Options.
• My models → Options has the model’s log, or see ~/.wanna/logs/.

Replies are slow

Check My models: if the model says part of it runs on the processor, choose a smaller version, less memory for conversation, or a model rated Runs great here.

A download stopped

Downloads resume where they left off, including after a restart. If one stays stuck, cancel it and start it again.

WSL2 stops suddenly while a big model runs

WSL2 ran out of memory. Use models that fit on your graphics card, or give WSL2 more memory in .wslconfig (Microsoft’s guide).

Still stuck?

Open an issue on GitHub with what you did, what happened, and the output of the system check.