Guides

Server & flags

Run llama-server without batch files and edit its flags through a form.

Start and stop

LlamaControl launches llama-server directly with an argument list — no shell, no quoting. Large models take a few minutes to load; the startup dialog closes itself when the model is ready and can be cancelled.

Editing flags

Flags (e.g. KV cache quantization, speculative decoding, fit-to-VRAM, context size, -ngl) are set in a parameter/value table. You can also paste a ready flag line, e.g. -ngl 999 --ctx-size 8192 --flash-attn on. Flags are saved per model.

Saving while the server runs restarts it immediately with the new config
Profiles — save flag sets under a name and switch between them

Live visibility

You see the server console log (save or clear it), CPU/GPU/RAM stats, VRAM usage, tokens per second, response time and uptime. The “Quick test” box lets you send a prompt without opening chat.

Engines

Models run on llama.cpp by default. An experimental KVMem engine is also available, installed into its own directory from Settings.