# ComfyUI LLM Party

A set of ComfyUI nodes for building LLM-driven workflows visually: multi-tool agent calls, role-based assistants, local knowledge-base RAG and GraphRAG, multi-agent interaction patterns, chat-platform integrations, and speech and image generation nodes in one graph.

ComfyUI LLM Party is a node package that brings LLM workflow construction into ComfyUI, so a user builds an AI assistant, an agent pipeline, or a full application visually and can combine it with existing image-generation workflows in the same graph. It adapts to LLM providers that expose an OpenAI-style or aisuite-style interface, including OpenAI, Gemini, Grok, Qwen, GLM, DeepSeek, Kimi and Doubao, as well as local models, GGUF files and vision-language models such as Llama-3.2-Vision, Qwen2.5-VL and Janus-Pro.

Nodes cover multi-tool agent calls and role-based assistant setup, retrieval-augmented generation with industry-specific word vectors and localized GraphRAG knowledge bases, single-agent pipelines as well as more complex multi-agent interaction topologies, integration with QQ, Feishu and Discord, and a combined LLM, text-to-speech and ComfyUI pipeline aimed at streaming media production. A bundled MCP server, and nodes for Omost, GPT-SoVITS, ChatTTS and GOT-OCR2.0, extend what a workflow can call.

It installs into an existing ComfyUI setup through ComfyUI-Manager, or as a Windows portable package that bundles ComfyUI with the plugin already included.

Website: https://github.com/heshengtao/comfyui_LLM_party