
Buying Computers for AI to Work On: 3 Mac Minis, 2 Phones, and the Ambition of AI Working Across Devices on Its Own
Vũ Khôi
Tác giả tại Wotech Electronics Blog

Buying Computers for AI to Work On: 3 Mac Minis, 2 Phones, and the Ambition of AI Working Across Devices on Its Own
A user shared a personal AI system of 3 Mac Minis, a Xiaomi 14 Pro, and an iPhone 17 Pro Max, via a write-up from the Hỏi Dân IT Official page. The interesting part, according to the write-up, is not the number of devices, but how they are connected so AI can work across machines — and one commenter argued the next trend is equipping dedicated computers for AI, instead of just buying a ChatGPT or Claude subscription.
Quick answer: The three Mac Minis each have their own job: the Mac Mini M4 runs services for outside users, connected via Cloudflare Tunnel and Tailscale so no port needs to be opened directly to the Internet; the Mac Mini M6 is the studio's main machine, running video editing tools, Hermes Agent, and self-developed AI programs, with both phones connected to it so AI can control them through Pinky; the Mac Mini M2 is the machine for testing software and new tools. All three machines are connected by Thunderbolt. Instead of sitting at a chat window issuing commands, they want AI to access the devices it needs to finish the work itself — for example, AI writes code on the Mac, accesses the Windows machine itself to run it, finds the error, fixes it, and rechecks, instead of the user acting as the middleman shuttling error messages between machines.
Cloudflare Tunnel and Tailscale are two ways to expose a machine on a local network to the Internet without opening a port directly on the router: Cloudflare Tunnel creates an encrypted tunnel from the machine to Cloudflare's infrastructure, while Tailscale creates a private virtual network (VPN) where only joined devices can see each other. Services on the machine remain reachable remotely while the attack surface exposed to the public Internet shrinks.
How are the three Mac Minis divided up?
According to the write-up, each machine has its own role:
| Device | Role in the system |
|---|---|
| Mac Mini M4 | Runs services for outside users; connected via Cloudflare Tunnel and Tailscale, no direct port open to the Internet |
| Mac Mini M6 | The studio's main machine: video editing tools, Hermes Agent, and self-developed AI programs; both phones connect to it for AI control via Pinky |
| Mac Mini M2 | The machine for testing software and new tools |
| Xiaomi 14 Pro + iPhone 17 Pro Max | Connected to the Mac Mini M6, for AI to control through Pinky |
All three machines are connected by Thunderbolt, making access and working back and forth more convenient. This division separates three concerns: stable services for outsiders (M4), creative work and the main agent (M6), and experimentation — where anything can break without affecting the other two machines (M2).
Why buy dedicated machines for AI?
One commenter in the write-up argued the next trend is equipping dedicated computers for AI, instead of just buying a ChatGPT or Claude subscription. They gave an example from their own system: Hermes Agent running on a server, a spare Android connected for AI to operate apps, and the main iPhone using TimiBot to assign tasks.
The argument here is not that hardware replaces the model, but that hardware gives the model a "body": instead of AI existing only inside a chat window, it has machines to run on, phones to operate, and a private network where the devices can see each other. A subscription buys access to intelligence; dedicated machines give that intelligence the ability to act on the user's own devices.
How is AI working across devices on its own different from the current way?
The write-up gives the example of developing an application: AI writes code on the Mac, but the app errors when run on Windows.
| The usual way | The way they want | |
|---|---|---|
| Writing code | AI writes the code | AI writes the code |
| Running it | The user carries it to Windows and runs it | AI accesses the Windows machine itself and runs it |
| Finding the error | The user encounters the error | AI finds the error itself |
| Fixing the error | The user copies the error message back to AI, AI fixes the code | AI fixes it itself |
| Rechecking | The user keeps checking | AI rechecks itself |
The user no longer has to act as the middleman shuttling error messages between machines. That is the core difference the write-up wants to stress: not AI writing better code, but AI walking the whole write — run — fix — recheck loop on the very device where the error happens, instead of stopping at the chat window and waiting for a human to bridge the gap.
Frequently asked questions
1. What does the personal AI system in the write-up consist of? According to the write-up, the system has 3 Mac Minis (M4, M6, M2), a Xiaomi 14 Pro, and an iPhone 17 Pro Max, with the three Macs connected by Thunderbolt.
2. How are the three Mac Minis divided up? The Mac Mini M4 runs services for outside users via Cloudflare Tunnel and Tailscale; the Mac Mini M6 is the studio's main machine, running video editing tools, Hermes Agent, and self-developed AI programs, and receives both phones connected via Pinky; the Mac Mini M2 is for testing software and new tools.
3. What does "buying computers for AI" mean? According to a commenter in the write-up, it is the trend of equipping dedicated computers for AI to work on, instead of just buying a ChatGPT or Claude subscription — so AI can access the devices it needs and finish the work itself, rather than existing only inside a chat window.
4. What roles do Hermes Agent, Pinky, and TimiBot play in the write-up? According to the write-up, Hermes Agent runs on the main machine/server to coordinate AI's work; Pinky is the bridge for AI to control phones; TimiBot on the main iPhone is used to assign tasks to AI. These are tools named in the write-up, not independently verified by Wotech.
5. What is the biggest benefit of this approach? According to the write-up, the user no longer has to act as the middleman shuttling error messages between machines: AI accesses the right device itself, runs the code, finds errors, fixes them, and rechecks, completing the whole loop without a human bridge.
Source: write-up by the Hỏi Dân IT Official page (18/11/2026); credit: Kan Kan. The configurations and interpretations in this article belong to the user and the commenter quoted by the write-up, and have not been independently verified by Wotech.
Bài viết liên quan

18 tháng 11, 2026 · 6 phut doc
FPT Software Chairwoman: US Clients No Longer Want to Pay by the Hour, Company Shifts to Outcome-Based Pricing
Artificial intelligence is forcing FPT Software to transform its traditional software outsourcing model: large clients in the US now apply 100% outcome-based payment, and the company expects the Time…

18 tháng 11, 2026 · 6 phut doc
Meta Expands Muse to Australia: The 6th Country, Real-Time Voice, and Local App Integrations
On 17/11/2026, Meta expanded Muse to Australia — the 6th officially supported country, just over 2 months after its US launch. According to a post by Hồ Hoàng Tiến (18/11/2026), the notable point is …

18 tháng 11, 2026 · 7 phut doc
AI Agent Costs Grow With the Square of Steps: 4 Layers to Cut Token Spend Before Switching to a Cheaper Model
You don't pay for answers — you pay for how many times you re-ask the same thing. That is the central argument of Hoang Viet AI in a post in the comailo group (18/11/2026): an agent running 40 steps,…