Troubleshooting

The model will not load

Read the red panel under the model bar. It includes the reason and the last llama.cpp error lines for that load. Choose View logs for the full runtime output. The same panel appears if the runtime stops after loading.

• Check that your runtime supports the model. Liquid d1-2B requires a build containing PR #31275, such as b11954 or a compatible newer build.
• If libraries are missing, set Shared-library folders in Runtime.
• To try CPU inference, set GPU layers to 0, then load again.

Use the error text to choose the next step. A Hub memory-fit estimate is not a loading guarantee. See Runtime.

A model download fails

Check that hf is installed and available. For a gated repository, confirm that your Hugging Face account has access and Noullama has an existing login or session token.

Only one model download can run at a time. A cancelled download keeps partial files for an immediate retry; quitting Noullama deletes those partial files. See download progress and retry.

Images are not available

Confirm the model supports images and that its compatible mmproj file is in the same folder. Enable Vision and load the model. Selecting a model from another folder turns the projector off.

For API requests, also check the 16 MB gateway request limit, including base64 overhead.

My client cannot connect

Keep Noullama running and use the endpoint shown in Local API. The port may have changed if 11435 was occupied. Do not connect to the private runtime port 11436.

Use localhost or 127.0.0.1 on the active gateway port. Foreign browser origins are rejected. The OpenAI adapter expects a JSON decision request in one user message, not an arbitrary chat prompt.

Closing the window asks to quit

Linux tray mode needs an AppIndicator library and a desktop tray host. GNOME setups may need an AppIndicator extension. Without tray support, Noullama asks before quitting when you close its window.

With a working tray, use Open Noullama to restore the window and Quit Noullama to stop the app and its local services.

There is no self-update option

Versions before 0.2.1 do not have an updater. Install a 0.2.1 or later .deb or .rpm once when available. Source builds and standalone executables require manual updates even when they show the newest version. See Updates.

Find your data

Settings, named presets and downloads use the platform app-data directory. On a typical Linux desktop:

~/.local/share/dev.noullama.desktop

Existing GGUFs are used in place. Changing the model download directory affects new downloads and does not move your existing files.

Report a problem

Include the installed version, distribution, selected runtime, model and relevant error lines when reporting a bug. Leave tokens and private state out of the report. For new capabilities, request a feature.