I Gave AI Its Own Computer (You Should Too)

I gave AI its own computer: a headless Mac Studio in my basement that runs ChatGPT, Claude, Grok Bot, Muse, Blender, Unreal Engine and local AI around the clock, completely separate from my personal files and logins.

I walk through the full setup: the computer-use settings for each AI app, what it all costs per month, the 3D and game tools the AI drives on its own, and how ChatGPT tuned a local Qwen model to 65 tokens per second, more than double the ~30 this machine normally manages.

Part 1: Cloud models. Remote access, monthly costs, and the computer-use and permission settings for ChatGPT, Claude, Grok Bot and Muse.

Part 2: Tools. Blender, Unreal Engine, Unity and Godot, with live demos of Claude and ChatGPT controlling them.

Part 3: Local AI. Hermes Agent, Qwen3.8 27B, a custom Android app, and fully local image generation with ComfyUI.

What would you have a dedicated AI computer work on? Let me know in the comments.

Setup

• Local model: Qwen3.8 27B-Uncensored-oQ4e-mtp, served with oMLX and used through Hermes Agent
• Remote access: Jump Desktop

What I pay each month

• ChatGPT 5x plan: $100
• Claude 5x plan: $100
• Grok Bot: about $30
• Muse: $16 (upgrade in the web browser, not the app, to save money)

That’s about $246/month total. You only need one of the $100 plans to get started.

Tools in this video

ChatGPT / Codex, Claude / Claude Code, Grok Bot, Muse, Jump Desktop, Blender, Unreal Engine, Unity, Godot, Hermes Agent, oMLX, ComfyUI, and Z-Image Turbo.

Links

Some links below are affiliate links. As an Amazon Associate I earn from qualifying purchases.

• Meta Wonder in the Muse App
• Opus 5.5 Built Skyrim in 3 Days
• GPT 6 Astra Builds a 3D Game in Blender + Unity
• Hermes + ComfyUI Image Generation
• X
• Website
• Mac Studio M4 Max 128GB, my dedicated AI computer
• Mac Mini, a cheaper way to start with cloud models
• Nvidia RTX 5090 (32GB VRAM)
• More cool tech on Amazon

Chapters

• Home AI Server Setup (0s)
• Mac Studio M4 Max 128GB for AI (73s)
• AI computer hardware requirements (112s)
• Part 1: Cloud AI models + Jump Desktop remote access (170s)
• ChatGPT vs Claude vs Grok vs Muse (219s)
• AI subscription costs per month (259s)
• ChatGPT computer use setup (375s)
• Claude computer use setup (510s)
• Grok Bot computer settings (647s)
• Muse computer use setup (748s)
• ChatGPT computer use demo (Godot game) (847s)
• Part 2: Blender, Unreal Engine, Unity and Godot with AI (972s)
• Claude computer use in Unreal Engine (live demo) (1068s)
• AI-generated assets: Blender to Unity (1273s)
• Part 3: Local AI on Mac Studio (1352s)
• Best local LLM for a 128GB Mac (Qwen3.8 27B) (1386s)
• Hermes Agent explained (1415s)
• Qwen3.8 27B at 65 tokens/sec (oMLX + MTP) (1569s)
• Hermes Agent Android app (1742s)
• Uncensored and abliterated local models (1821s)
• Local AI coding test: Asteroids game (1917s)
• ComfyUI local image generation with Hermes (2042s)
• Z-Image Turbo results (2220s)
• Final thoughts (2346s)

Watch on YouTube.