NodeGPT

When one agent just isn't enough, jam two together

The Assistant node is your default agent — AutoGen's workhorse, with a name you set

The Chat node is where the conversation actually happens — NEW to start, SEND to continue

The 'ChatGPT' node that isn't a chatbot — it's the brain box behind your AutoGen agents

The node that hands your LLM's answer to Stable Diffusion — a CLIPTextEncode with extra steps

Your AI's designated contrarian

Finally, a way to actually see what your agent said

When the graph needs your LLM's words as pixels

The Engineer agent writes the code; don't let it run anything

The Executor agent runs the code the Engineer wrote — treat it with respect

Round up a whole squad of agents and let one referee

Run a GGUF model right inside the graph — slowly, but locally

The node that gives your workflow eyes — image in, caption out, fully local

The pack author's own recommendation for running local models

The node that keeps your API key out of your workflow JSON (listen to it)

Drag a text file into your graph like it's any other asset

Local models through a LiteLLM proxy, and the defaults are a trap

Point NodeGPT at text-generation-webui's OpenAI-compatible API

The type adapter that lets your agent talk to the rest of ComfyUI

Think before you act, in agent form

The code-writing AutoGen agent hiding inside ComfyUI

NodeGPT's bare-bones single-shot LLM call

How NodeGPT saves the whole agent conversation

The agent that actually runs the code in NodeGPT