What Is Comfy Agent? An Agent That Builds ComfyUI Nodes

ComfyUI announced Comfy Agent on October 1, 2026. ComfyUI is one of the most popular node-based interfaces for running image, video and audio models. At first glance it looks full of nodes and noodles and hard to learn. Comfy Agent aims to lower that barrier: you write what you want, and the agent plans, builds and runs the workflow.

This article summarizes the Comfy Agent section of the AI Search channel's weekly AI news video on YouTube and extends it with Comfy's official announcement; the video link is at the end. This is a news roundup, and I have not tested the feature. Information comes from Comfy's announcement page; the product is in beta and may change.

What is ComfyUI, and why an agent?

To produce an image or video in ComfyUI you connect models, samplers, extra modules and inputs with nodes; that node graph is called a workflow. The flexibility is great, but building a workflow from scratch or repairing a broken one means knowing which model, which file and which node is needed. Comfy Agent is designed to take over exactly that knowledge burden.

What does Comfy Agent do?

According to the announcement, it is an agent built into ComfyUI that "plans, builds and runs" what you describe. Details:

• Model selection and workflow building: It picks and connects the right model and nodes for the result you want. For example, if you say "generate a video with MiniMax," it can find the right workflow and run it (the video's example).
• Error fixing: If you hit an error message, you ask the agent right inside ComfyUI.
• Asset discovery and jargon translation: It translates technical terms into plain language and finds the models and files you need.
• Visual understanding: It can understand the content of images you give it.
• Live workflow view: You see the workflow the agent builds in real time.
• Parallel chats: Up to five conversations at once, with full chat history.
• Knowledge base and skills: It has access to the latest models, ComfyUI extensions and workflows; you can create and use public or private custom skills.

The example requests in the announcement show the agent does not only build from scratch but also improves existing work:

• "Update this same generation to compare 3 more video models."
• "Explain what this workflow does and where it's slow."
• "Re-run my product-shot workflow on these 20 images."

Announcement: blog.comfy.org/p/comfy-agent-the-first-agent-for-craft.

Where and how can you use it?

For now in Comfy Cloud (as of October 1, 2026), ComfyUI's online service. According to the announcement:

• Free to try: You can try it with your existing Comfy credits; opening the agent panel is enough to claim the initial free chats. After that, pricing is described in the docs as per token.
• Getting started: Go to cloud.comfy.org, open the "Agent" panel, claim the free starter chats and write your request in natural language.
• Desktop: It comes to Comfy Desktop "in a few weeks." So the version running on your own computer does not have it yet.
• Local installs: The announcement says they will later support local installations with GPU access and custom nodes.
• Status: Beta. More model options and external integrations are planned.

What does it mean for local workflows?

ComfyUI's strength is combining many models in one place; its weakness is the learning curve. If an agent builds and runs the workflow, the barrier drops for people who want to try open-source models. For example, installing LTX 2.5 with ComfyUI today takes a file list and folder structure; it is a natural expectation that some of those steps will be handed to the agent. But until the desktop version arrives, this is limited to workflows in the cloud service; the setup on your local GPU does not have the agent yet.

Practical tips:

1. Review the workflow it builds before you run it. Knowing which model gets downloaded and which nodes are used saves time when debugging.
2. Paste error messages to the agent. As the presenter in the video also points out, that is the use that helps most with technical problems.
3. Delegate comparison jobs. Repetitive work such as "run the same prompt on 3 models" suits an agent.

For the same week's image models, see Ideogram 4.5 vs Flux 3 Image; for methods that speed up video generation, see the DMAD and PDMD article.

Frequently Asked Questions

Is Comfy Agent free?

You can try it for free in Comfy Cloud with existing credits, and initial chats are provided. Afterwards, the docs describe a per-token pricing model.

Does Comfy Agent run locally?

Not yet. It is expected to come to Comfy Desktop within a few weeks; support for local installs and custom nodes is planned for later.

What does Comfy Agent do?

It plans your request, picks the model, builds and connects nodes, runs the workflow, tries to fix error messages and explains or edits existing workflows.

How many chats can I open at once?

According to the announcement, up to five simultaneous conversations with full history.

What is ComfyUI?

A node-based interface for running image, video and audio models.

Source

• AI Search, Gemini 4, GPT 6.1, Dots, Claude Sonnet 5.5, Ideogram 4.5, Flux 3: AI NEWS (YouTube, October 4, 2026): the Comfy Agent section.
• Comfy, Comfy Agent: The first agent for craft (October 1, 2026): features, access, pricing and status.