Agentic Autonomy Building a Local Dual Server AI in ComfyUI

In this guide, we’re moving past the "AI as a toaster" phase and building a truly autonomous, multi-agent system entirely on local hardware. No cloud APIs, no subscriptions—just pure agentic power running in a dual-server setup within ComfyUI.

We’re pairing a lightweight Ollama (Phi-4-Mini) "Director" for logic and instructions with a heavyweight vLLM (Gemma) "Executive" for high-fidelity output. By splitting the cognitive workload, we eliminate hardware bottlenecks and create a private AI studio that works for you, 24/7.

What we cover:
* The architecture of a local Multi-Agent system.
* Setting up the foundation: WSL, Python Venv, and vLLM.
* Configuring the "Director" in Ollama with custom ModelFiles.
* Building the "Semantic Bridge" in ComfyUI.
* Optimizing for 16GB-24GB VRAM hardware.

Resources:
* Guide: installing vLLM and ComfyUI in wsl - https://zanno.se/guide-run-gemma-4-nvfp4-from-comfyui/
* Guide: Running a 3 server agentic AI inside ComfyUI - https://zanno.se/guide-setup-local-agentic-ai-in-comfyui/
* GitHub: https://github.com/Creepybits/ComfyUI-vLLM-MultiModal-Agent
* Join the newsletter: https://zanno.se/newsletter/

⏳ Timestamps
0:00 – The Dream of Local Autonomy
0:28 – The Dual-Server Architecture (Director vs. Executive)
0:42 – Software Prerequisites (WSL & Virtual Environments)
1:05 – Configuring the Director (Ollama & Phi-4-Mini)
1:54 – Creating the ModelFile & Custom Parameters
2:27 – Organizing Your Workspace: Folders & Workflows
2:53 – Installing the Multi-Modal Agent Node Pack
3:02 – Setting up the Background API & Data Routing
3:32 – The Startup Sequence: Powering up the Servers
3:52 – Running Your First Task in ComfyUI
4:05 – Analyzing the Output: Logic to Execution
5:05 – Beyond the Cloud: Your Private AI Studio

Видео Agentic Autonomy Building a Local Dual Server AI in ComfyUI канала Creepybits