Boltjar and ComfyUI

Boltjar owes ComfyUI its core idea: a typed node graph, where every input and output has a type and you rewire the whole pipeline by hand. The two tools point that idea at different jobs.

What each is for

ComfyUI is a node-based engine for generative media. You build a workflow that makes images, video, audio, 3D or text with models such as diffusion models, queue it, and it runs through once and produces its outputs. It re-runs only the parts of the graph that changed since the last run, which suits trying one change at a time. It is licensed GPL-3.0.

Boltjar is a node editor for systems that stay on. A graph is a server: you turn it On, and its triggers (chat messages, webhooks, schedules, timers) keep firing it until you turn it Off. Its nodes lean toward chat models, tool calls, speech, stores and HTTP rather than image generation. It is licensed AGPL-3.0 with a node pack exception.

Side by side

	ComfyUI	Boltjar
A run	a queued job that runs once	a graph that stays On and runs on every trigger
Starts from	the queue button or the API	chat, webhook, schedule, interval, manual, data
Main media	images, video, audio, 3D, text	text, speech, images for models that read them
Models	runs models in its own process	calls Ollama or a provider’s API
Wires	typed	typed: events push, data is pulled
Custom nodes	Python custom nodes	Python node packs

Which one to use

Use ComfyUI to generate media and iterate on it. Use Boltjar for something that has to keep running and respond: an assistant, an endpoint, a scheduled job, an agent with tools. They also work together: a Boltjar HTTP Request node can call a running ComfyUI server’s API.