Repository navigation
When something doesn t work
overcatbe edited this page Oct 25, 2026
·
1 revision
- A model doesn't load: the pane at the bottom shows the last lines of llama-server's log. "Too big" in the Memory line means it doesn't fit the graphics card: lower Context size, tick Compress context, or set On the GPU to auto.
-
"No answer." under the chat: llama-server stopped while answering. The note says why,
from its log. A picture can be more than the model takes in one batch (Gemma models crash on
that): models with eyes get a batch of 2048 by themselves; for bigger pictures add
-b 4096 -ub 4096to the launcher's options. This is a model setting (Launchers tab), not a persona one: it is fixed when llama-server starts. - A tool's terminal says "must be run from a terminal" or can't ask for a password: tick Start and stop in a terminal window under Edit….
- A persona doesn't use the files: check that it has Files set (and saved) on the Personas tab, and that the folder is allowed on the Settings tab. Its model has to know tools: the Launchers tab says can use files or can't use files next to the file size, and the Personas tab warns when a persona with Files runs on a model that can't. If the model answers "I would read…" without doing it, it is not good with tools; try another one.
- A keyboard button does nothing: the pane at the bottom shows what its command answered. "Error loading new keyboard description" means the system doesn't have that layout or variant; see the topic Keyboard.
- Passwords: "The key of this store is not on this machine": More → Import key, with the key file exported on the machine that has it.
- Anything about paths or programs: Settings tab, Check.
Made from the guide inside AIcove (cove/data/guide.md); it opens with the ? button.