Nodes/NodeGPT/Planner
ComfyUI Node

Planner

Think before you act, in agent form

By antonym-git·Created 3 years ago·Updated 2 years ago· 355
Planner
  • LLM
  • Agent
Seed42
Temp0
request_timeout120

The Planner is the strategic thinker in NodeGPT's agent lineup. Where the Engineer writes code and the Executor runs it, the Planner's job is the step before either: break a task down, decide the order of operations, and hand the other agents something structured to execute. It's the difference between "write a script that draws a spiral" as a vague demand and as a numbered plan the team can follow.

It's the same AutoGen machinery as every agent node in the pack - you give it an LLM, it gives you an Agent - with a system message tuned for planning rather than doing. Put one in a GroupChat alongside an Engineer and a Critic and you've got the classic "plan, build, review" pipeline that the pack's whole multi-agent feature was pointing at.

Inputs and output

  • LLM - required, from any provider (ChatGPT, LM_Studio, Ollama, llama-cpp, oobaboogaOpenAI).
  • Seed / Temp - default 0. For a planner you mostly want determinism; a wild plan is usually a bad plan.
  • request_timeout - 120s default; planning long tasks can run long, so expect to raise it.

Output: Agent, into a GroupChat or Chat.

The honest appraisal

The planner-actor split is one of the best patterns AutoGen popularized, and it's aged better than this pack did. The implementation, though, is the usual NodeGPT story: unfinished, with community reports of agents echoing input and group chats that wouldn't stop until the queue was killed. The author's own README says "not everything is functioning correctly yet," and the repo is now gone entirely - xXAdonesXx/NodeGPT 404s as of this writing, the GitHub account is deleted, and the README's install is historical:

cd ComfyUI/custom_nodes
git clone https://github.com/xXAdonesXx/NodeGPT

Restart to auto-install requirements (including pyautogen). If you're building multi-agent workflows in 2026, use a maintained framework and keep the division of labor - it's the part of this pack worth stealing. And if you find a Planner in an old saved workflow, you now know exactly what it's supposed to do: think first, so the doers have something to do.

CategoryAutoGen/Agents

Inputs (4)

NameTypeDefaultDescription
LLMLLM
SeedoptINT42
TempoptINT0
request_timeoutoptINT120

Outputs (1)

NameTypeDescription
AgentAgent