LM Studio

Of all the local-LLM options in NodeGPT, this is the one the author actually vouched for. The README's "Usage" section is two lines long: start the LM Studio server, start ComfyUI, place nodes. That's the entire recommended path for a fully local, no-API-key setup - and it tells you the LM Studio node is the one the author himself reached for.

LM Studio is a desktop app that downloads and runs GGUF models locally and exposes them over an OpenAI-compatible API. That last part is the whole trick: because it speaks the OpenAI protocol, a node can treat it like any other LLM backend. That's why this node has so little on it - all the complexity lives in LM Studio, and the node just points at it.

The two fields

• api_base - defaults to http://localhost:1234/v1. That's LM Studio's default server address, which is how you know this node is a thin client: the default is already correct for a stock LM Studio install. Change it only if you're running the server on another machine.
• api_key - defaults to the string NULL. This is the tell that no real key is involved: LM Studio is local, so any value (or none) satisfies the OpenAI-shaped handshake. Don't let the field's existence make you think you need a cloud key.

Output: LLM, feeding any agent node - Assistant, Engineer, Critic, Planner, Executor, or GroupChat.

The workflow

# in LM Studio: load a model, then start the local server (port 1234)
cd ComfyUI/custom_nodes
git clone https://github.com/xXAdonesXx/NodeGPT

Restart ComfyUI to auto-install requirements, wire LM_Studio → Assistant → Chat, and you've got a fully local agent chat. Community users at the time confirmed this combination worked - LM Studio was genuinely the smoothest local backend for the pack.

What to expect

The usual pack caveats apply: chats that don't terminate cleanly, agents that sometimes echo their input, and an author who called the whole thing unfinished. The repo is also now gone - xXAdonesXx/NodeGPT 404s as of this writing and the GitHub account is deleted, so the clone above is historical. That's a shame, because the integration pattern aged beautifully: "local app speaks OpenAI protocol, node points at it, done" is exactly how modern ComfyUI LLM packs still connect to LM Studio and its ilk. If you take one thing from this node, take the port number - 1234 - and the realization that the easiest local-LLM path in ComfyUI has barely changed since 2023.