skills.rest
artokun
artokunCOMMUNITY
Explore all skills by this author →
Website𝕏@artokunGitHub

local-llm-free

Configure Ollama with fine-tuned Gemma 4 models to run the ComfyUI agent offline.

715|111|Updated Feb 15, 2026
Search Tags:#offline#ollama#free#ComfyUI#Gemma#local LLM#fine-tuned model
One-click install
npx skills add https://github.com/artokun/comfyui-mcp --skill local-llm-free
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: local-llm-free
Source: https://github.com/artokun/comfyui-mcp/tree/main/plugin/skills/local-llm-free
Command: npx skills add https://github.com/artokun/comfyui-mcp --skill local-llm-free

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Running an AI agent against ComfyUI normally requires a paid subscription or a hosted API key. This Skill shows how to run the full comfyui-mcp agent locally and offline for free using Ollama and Gemma 4 models fine-tuned on the comfyui-mcp tool suite.

Core Features & Use Cases

  • Ollama Setup Guidance: Step-by-step instructions to install Ollama and pull the right fine-tuned model variant (:e2b, :e4b, or :12b) based on available GPU VRAM.
  • Model Selection by VRAM: A sizing table maps 2 GB, 4-7 GB, and 8 GB+ VRAM budgets to the appropriate model rung, with arena benchmark scores for each.
  • Expectation Setting: Documents known limitations such as no vision capability, audio input restrictions, cold-start latency, and compact tool mode for non-panel MCP clients.
  • Use Case: A user with a 6 GB GPU and no API budget asks to run ComfyUI generation offline; the Skill directs them to pull artokun/gemma4-comfyui-mcp:e4b and connect via the panel's Ollama backend.

Quick Start

Ask the agent to set up the free local Ollama backend with the recommended fine-tuned Gemma 4 model for your GPU's VRAM.

Frequently Asked Questions about local-llm-free

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run the ComfyUI agent locally for free without an API key?▼

Install Ollama, pull the fine-tuned model artokun/gemma4-comfyui-mcp:e4b, then select the Ollama (local) backend in the ComfyUI sidebar panel and connect. No subscription or API key is required, and it works fully offline.

Which local Ollama model should I pick for my GPU VRAM?▼

For 2-3 GB VRAM use :e2b, for 4-7 GB use :e4b (the default and best local arena score at 14/20), and for 8 GB or more use :12b for steadier multi-step tasks. Match the rung to your free VRAM.

Can local Ollama models analyze images or audio in ComfyUI?▼

The fine-tuned Gemma 4 models keep tool calling but have limited or no vision, so they cannot visually critique outputs. Audio input is refused unless the model is in the verified set (gemma4:e2b, gemma4:e4b, nemotron3:33b).

Why is the first request to the local model so slow?▼

The first request after connecting triggers a cold model load into VRAM, which can take 30 seconds or more. This is normal behavior for Ollama and subsequent requests run much faster.

Can I use these fine-tuned models with other MCP clients besides the panel?▼

Yes, any Ollama-speaking MCP harness such as Hermes or OpenClaw can use these models. Pair them with compact tool mode via the --compact flag for reliable tool calling outside the panel.

Related Skills

Explore More Software Engineering Skills→
video-production-buddyvideo-production-buddy
602

comfyui

Configure ComfyUI workflows for image and video generation.

Official
Intermediate
519lab519lab
2

comfyui

Generate and manage images, videos, and audio with ComfyUI workflows.

Official
Advanced
wshobsonwshobson
39.3k

checkpoint-promotion

Gate fine-tuned checkpoints through drift budgets, paired arenas, and canary rollouts before promotion.

Community
Advanced
zeketxzeketx

openai-agents-sdk

Create and manage AI agents with OpenAI Agents SDK, Gemini, and OpenRouter.

Community
Advanced