TL;DR
Creator Mike Russell unboxes two Apple Mac Mini M4s (16GB RAM, 256GB SSD) and live-streams setting them up as dedicated local AI servers on his home network. One Mac Mini is configured to run local AI models via LM Studio and Parakeet MLX for transcription, while the other acts as a headless web-browser controller for AI agents to scrape real web data through logged-in and logged-out Chrome profiles.
Key points
1
Two Mac Mini M4s (16GB RAM, 256GB SSD) are being set up as headless local AI inference servers alongside an existing Mac Studio M3 Ultra, forming a three-tier local AI stack
2
Essential macOS settings for a headless AI server: enable Screen Sharing, Remote Login, and File Sharing; set display and password lock to Never; enable auto-login and startup after power failure
3
One Mac Mini is configured to run Qwen 3.5 9B (4-bit MLX, ~5.98GB) via LM Studio for text inference and Parakeet MLX for local transcription, handling lightweight grunt-work tasks
4
A second (older M2) Mac Mini runs dedicated Chrome browser instances, both logged-in and logged-out, allowing AI agents (via Chrome DevTools MCP) to browse the web as a human user and extract live data
5
Tank, the creator's AI coding orchestration system, drives all Mac Minis via SSH and natural language commands, automating model downloads, LM Studio config, and live browser control
Actionable insights
→
To run a Mac Mini headlessly: enable Screen Sharing, Remote Login, and File Sharing in System Settings; set lock screen and password to Never; enable auto-login and after-power-failure restart so the machine is always ready without manual intervention
→
For a 16GB M4 Mac Mini, Qwen 3.5 9B in 4-bit MLX format (~5.98GB) is a practical text model that leaves enough headroom for the OS, while Parakeet MLX handles local audio transcription without LM Studio
→
Dedicate a separate low-cost Mac Mini purely to running Chrome browser instances (logged-in and logged-out profiles) so AI agents can retrieve paywalled, login-required, or JavaScript-heavy web data that standard API tools cannot access
→
Use the native macOS Screen Sharing app and SMB file sharing to manage multiple headless Macs from one main machine without any third-party software
→
Reserve the large Mac Studio M3 Ultra for 70B+ parameter model inference, and offload categorisation, tagging, yes/no decisions, and transcription to smaller Mac Mini M4 models to avoid wasting high-end compute
Notable quotes
“I always have now as a backup in my system prompt: if you are struggling to get some web data from somewhere, go ahead and drive the browsers on this Mac Mini.”
“I find scraping is like a dirty word, isn't it? I would far rather say extracting than scraping.”
“I am like, so local, it is unreal. And I love this.”
Worth watching?
⏭️
Worth watching the full video?
The setup instructions and key configurations are all captured here; the video itself is a loosely structured live stream with significant dead time waiting for downloads, live stream troubleshooting, and off-topic pizza and coffee chat, so reading this summary is the more efficient option.
Saved you some time? The creator still deserves a like.
Found this by searching? That's the hard way.
summree watches the AI channels you follow and sends you the summary the moment a video drops. Nothing gets past you. From £4/month.
Try it free