Local AI Status

Local AI Status is a pure observability node: it reports what the pack's managed llama-server process is doing, in one status string. It takes no inputs and is an output node, so you drop it on the canvas and read the preview. If you're on the default release_after_generate = true setup there's rarely a server running to look at - this node earns its place the moment you flip to keep-alive mode or set release_after_generate = false and a server sits resident between calls.

What you get in the status output, all in one line:

• PID - the actual OS process id of llama-server.
• model - the filename of the loaded GGUF, so you can confirm which model is actually resident.
• endpoint - the localhost base URL and port the server is serving on.
• healthy - a live /health check against the server.
• modalities - which capabilities the server advertised (text, plus image and/or video when a compatible projector is loaded). Seeing text only is your first clue the projector didn't take.
• keep_alive - seconds remaining until an idle server auto-releases, or manual if you set keep_alive_seconds to zero.

It's a thin node, and that's fine - it does one thing well. The node forces a re-run on every graph execution (its IS_CHANGED always returns changed), so the status you read is current rather than cached from the last run.

The genuinely useful workflow: run it before and after Unload Local AI Model. Before, it confirms the server is resident, healthy, and how long until it self-releases; after, it should report "No managed llama-server process is running." That's the cheap way to confirm you got your VRAM back before a heavy diffusion or video generation - the exact sequence the pack's README tells you to do manually, now with proof.

The limitation is worth stating plainly: it can only see the pack's own managed server. If you started llama-server yourself outside ComfyUI, this node knows nothing about it, and ComfyUI will still be blind to that process's VRAM. Use it to monitor the server the pack spawned, not llama.cpp in general.

Install is the pack standard - ComfyUI Manager, search WepeNerd, or git clone https://github.com/WepeNerd/ComfyUI-WepeNerd.git into custom_nodes plus pip install -r requirements.txt, restart. No model, no projector, no llama-server needed for this node to load - it only talks to a server that already exists.