The connection test passes but the model won’t reply
The connection test retrieves the provider’s model list. Select a model, save, and send a short message to test generation. Check the exact model name, access, and account balance if it fails.
The app or API key storage fails on Linux
AppImage needs FUSE 2. On Ubuntu 24.04 the package is
libfuse2t64. Make the file executable before opening
it.
Enable a system keyring such as GNOME Keyring or KWallet and restart codef if it cannot save an API key.
A download stalls or fails
Open Settings → Models and downloads → Downloads to check whether a file is queued, downloading, or importing. You can cancel or retry an individual download. Quitting codef stops active downloads.
Check free disk space. If access to the model is restricted, request access on Hugging Face and then add your Hugging Face token in Models and API.
A local model fails to start or runs slowly
Check that Ollama is installed and running on Windows or Linux. Try a smaller model or reduce the context size. Use a vision-capable model for photos and a tool-capable model for agent work or web search.
Where chats and attachments are stored
Open Settings → Data → Open data folder to find chat history and attachment copies. The agent accesses project files according to its selected permissions.
Check token usage
Settings → Usage shows token counts for codef requests. If a provider omits its count, codef estimates it. The indicator above the input shows context usage for the current chat.