Stop llama.cpp server

This is the mop-up node. "Stop llama.cpp server" does exactly one thing: tells the pack's server manager to shut down the llama-server process running on a given port. That's the entire job, and it's worth knowing when you actually need it - because for most workflows, the answer is never.

How it works. The node's only input is port (8080 by default). When it runs, it calls the pack's server manager, which terminates the process it's tracking for that port - a graceful terminate first, a hard kill after ten seconds if it doesn't go quietly. The output is a single status string that reads something like stopped server on port 8080. Wire it to a display node if you want proof; otherwise it's the kind of node you drop in, run once, and forget.

The honest limitation. It only stops servers this pack started. The pack keeps track of the llama-server processes it spawned in memory, and the stop node works on that registry. If you launched llama-server by hand in a terminal - which is a perfectly reasonable thing to do - this node won't touch it. You'll have to close it yourself. Killing on the wrong port is harmless, though: nothing happens, status still reports success.

When you'd actually use it. Three situations:

• You disabled release_after_use on the LLM node (so the server stays alive between calls) and now want VRAM back before the diffusion pass runs.
• A server got stuck - generation hung, port won't release - and you want a clean kill without restarting ComfyUI.
• You're debugging a port conflict on 8080 and want to make sure the pack's own server is gone before you investigate what else is squatting there.

The thing to notice is that for the default happy path you never see this node. auto_start launches the server on first use, release_after_use tears it down when generation finishes, and the pack even registers an exit hook so a stale server dies when ComfyUI itself exits. The lifecycle is managed for you; this node is the manual override that exists in case the automation is not what you want.

Install and a gotcha. Same pack-wide story - ComfyUI Manager, search "ComfyUI-llamacpp-helper", or cd ComfyUI/custom_nodes && git clone https://github.com/bbaudio-2025/ComfyUI-llamacpp-helper, then restart. No Python dependencies, nothing to pip install. The only real trap is the port: the stop node's default (8080) matches the Load node's default, but if you changed the port there, change it here too - stopping port 8080 while your model actually runs on 8081 is the most anticlimactic troubleshooting session you'll have all week.