Your local AI agent is actually locked up, three tools make it available everywhere
脈報 · 2026-06-01 · 9 min
Show full episode description
本地 AI agent 常被困在一台電腦上，人一離開就用不到。這篇用 Tailscale、Termius、tmux 三層工具，讓它隨處連得到、任務不中斷又不暴露公網，並點出連得到不等於安全。 ⭐ 文章深度讀：多數 agent 教學只教你裝功能，卻沒教你怎麼從手機隨時連到它、又不暴露公網 → https://heymaibao.com/local-ai-agent-remote-access/ ⚡ 章節重點 開場
📑 Chapters — tap a time to jump there
00:00
本地 agent 的困境
01:58
Tailscale 連得到
03:03
Termius 能操作
04:31
tmux 不中斷
05:46
連得到不等於安全
07:01
無聊基建的價值
🌐 This transcript was automatically translated to English from the original.
Hi, welcome to this content analysis. Imagine this scene. You painted for a whole weekend, bought a hand in front of the screen, taught your AI agent a bunch of awesome skills, installed various plugins, set up Chromejob, and even wrote several automated scripts that will run silently in the background. You watch it work perfectly, and you feel overwhelmed and have a lot of fun. This is really a thing that greatly improves productivity, right? But here comes the problem. When you leave that desk, will all this beauty still exist? So I want to ask you a super direct question, is your autonomous AI really autonomous? Many of us can actually only run the local agent on a fixed machine. It may be the most powerful desktop computer in your home, it may be a Mac mini that you stuffed in the corner, or it may be an old laptop that you transformed into a constantly open agent box. When you sit in front of it, the following are great, It looks smart and autonomous. However, as soon as you leave the house, if it can only be stuck in that machine, then to be honest, it is not practical at all. As the original author said a very bloody sentence, when you leave that chair, your agent is not autonomous at all. I believe many knowledge workers have experienced this kind of frustration. You started a research task that took several hours to run with great expectations, and then confidently closed the laptop and took the MRT. What happened? SSH was disconnected and changed to a WiFi network. Or you just want to use your phone to peek at the progress while commuting, but find that you can’t connect at all. At that moment, the so-called automation is really empty talk. It doesn’t matter, today we will take you to break this physical prison. Our blueprint is very clear. We will first show you the dilemma of local AI, and then introduce the three magic forces. Talescale allows you to connect, Termius makes it easy for you to operate, and Tmox ensures that tasks are not interrupted. Then we will clarify the boundary between connection and information security. Finally, we will introduce you to these boring infrastructures. What is the huge value? Okay, let’s go to the first part. The dilemma of local AI is the practical physical cage. What exactly do we want to achieve? In fact, it is very simple, just four wishes. First, we want to be able to access the agent from any place and any device. Second, for security reasons, we absolutely do not want to expose the connection department to dangerous public networks. Third, we do not want the kind of fragile connection that dies after a short period of time. Fourth, we do not want to restart the task every time we change the device to check the progress. To meet these four points at the same time without relying on a stronger AI model, what we need is actually a set of better and more secure infrastructure settings. In order to achieve this perfect state, we have built a trilogy of super-skilled new models. These are the three keys to liberating your trapped AI. Connectivity, operation, and non-interruption. As long as you get these three things, your agent can be responsible for what it should do without any worries, which is to help you improve productivity. Let's take a look at the first key, part 2, Tailscale. The key point is to get the connection, and it is a secure connection with zero public connections. Tailscale is a really amazing tool. It solves extremely complex network problems in an extremely simple way. The underlying technology is WireGuard, which can automatically help all your devices to establish a private mesh network. To put it bluntly, your desktop, laptop, mobile phone and even server will all get their own private location. They have a very funny name. They can have crazy conversations with each other in private. It is almost like they are all moved. It’s as magical as being in the same physical room. Why is this so important? Think about how we did it in the past. Public IP plus Power Forwarding on the router to forward it out. But God, you have to connect, but the AI Agent with the system command line is also directly exposed to the public network. It is basically opening the door to the name of hackers. So Tailscale’s zero public connection strategy is absolutely key. There is no public IP, no complicated router settings, and no SSH Port alone on the network waiting to be attacked. Your machine is completely invisible. By the way, there is a very critical technical detail here. The majority of Windows users must listen carefully. If your AI Agent is running in a WSL environment, remember that Tailscale must be installed in WSL and not just on your Windows host. Just ensure this small detail and your connection settings will be seamless. Then comes the third part of Termius. We must not only be able to connect but also operate. Put AI directly into your pocket. This step is to make your AI Agent Freed from that bulky desktop computer and turned into a portable tool. What do you think? Everyone knows how to use laptop SSH connection. But what if you are out today and only have a mobile phone at hand? To be honest, a lot of SSH Apps for mobile phones on the market are really difficult to use. The keyboard configuration is extraordinary and the connection is stuck. At this time, Termias has become our savior. The mobile phone board of Termias is designed to be super elegant and practical. You can see that it has an exclusive mid-range basic keyboard and complete SSH key management. The host interface is also clear at a glance. The craziest thing is that it supports connection forwarding. What does this mean? That is to say, you can set it up in Termias and then directly open Safari or Chrome on your phone to directly see the super gorgeous dashboard of your AI Agent. You only need to use the name given by Talescale to build the host, set the key, and connect with one click. The phone will instantly become the most powerful idle station. This trick is really awesome. Okay, let’s go to the fourth part. Tmox solves the problem of non-interruption, allowing you to run without interruption and in the background. This tool solves it perfectly. The pain point of standard SSH being too fragile is that you usually use SSH to connect and run an analysis task for several hours. Then you turn off the laptop or walk into a place with no signal on the MRT. Bang! Once SSH is disconnected, your task will be buried with it. It will really crash, right? But with T-MOX, the situation is completely different. Just let the connection be disconnected and your task will still live well on the remote machine and continue to run smoothly. You can imagine that Tmox is the boss who really owns and controls the command line. And SSH is just a window that you use to look in. Your Agent actually works in the safe room of T-MOX. So when you go to a place with poor signal, the SSH window is forced to close, so what. The T-MOX in the room is still happily directing the Agent to work. As long as the host is not turned off, you can always take another device to SSH in and connect to T-MOX. You will find that everyone is working on the same stage and continue to work hard. It is very important to come to the fifth part. Connection does not equal security. Let’s talk about the blind spots of full-line control. This is definitely the sentence that needs five stars in today’s analysis. The original author gave us a very honest reminder of reality. This combination protects how you connect to the Agent, not what the Agent can do after connecting. Don’t get confused. T-SKL protects the connection path. SSH gold medicine blocks uninvited guests. T-MOX ensures that progress is not interrupted. But all this has nothing to do with the behavioral authority of the Agent itself. Think about these potential risks. If your Agent is given too much authority, it can mess with your personal files, see API elixirs in the system environment variables, modify private program libraries, and even itself is vulnerable to prompt injection attacks, that is, the impact of Prompt Injection. Then this infrastructure does not perform any magic to make it safe. If you are dealing with sensitive data, you still have to set up the full line control within the system, and even study Telscale's ACL access control. To harden your security defenses. Finally, we come to the sixth part. The value of boring infrastructure, which is the dividing line between good and usable. We all know that the most fascinating and fun part of AI is those cool plug-ins and magical automation magic. But if it is constantly disconnected, or can only stay in one room, then it is just an expensive toy. Establishing a set of infrastructure that can operate without interruption, which sounds boring, is what makes your AI Agent. The only way to advance from fun to really usable. This may not be the only solution, but I think it is currently the most friendly to novices, the fastest and most practical chess player style. So in the end I want to leave this question to everyone to think about. Have you set up your boring infrastructure before crazy pursuit of those fun AI functions? I hope today’s analysis can help you re-examine your workflow and truly liberate the AI assistant trapped behind the screen. That’s it for today’s sharing. Hurry up and try it out. A combination of Talescale, Termias and Tmox. See you next time