Repository navigation
Settings
Everything that differs from one machine to the next: the llama-server program, your AIcove folder, the model folder, how much memory the graphics card has, how much of a model goes on it by default, the ports, the terminal program for tools, the keyboard buttons, what the AIs may read and write, the password store and the notes file.
- Find again looks the machine over and fills in what it finds.
- Check tries every entry and says what is wrong. It checks what is in the form, saved or not.
- Save writes the settings.
On a machine where AIcove runs for the first time, it opens on this tab with what it found.
Keyboard buttons: the buttons in the top bar that switch the keyboard layout. See the topic Keyboard.
Which llama-server: AIcove doesn't bring llama.cpp; install it as llama.cpp's own install guide describes, or take a ready build from its Releases page on GitHub. Pick the build for your graphics card: CUDA for NVIDIA, ROCm for AMD, SYCL for Intel, Vulkan for any of them (Vulkan is a standard that the normal drivers of all three speak; the vendor's own build is usually a bit faster), CPU without one. What fits: a model file needs about its own size in memory, plus 1–3 GB for the conversation; the Launchers tab shows the estimate for each launcher.
Model reachable from: "this machine only" is the safe choice. "Everyone on the network" lets other devices talk to the loaded model directly, without a password. Takes effect when the model is next started.
Made from the guide inside AIcove (cove/data/guide.md); it opens with the ? button.