Stop messing with CLI locally: Why should I build a "cloud multi-agent hub" by myself?

Recently, terminal-based AI agents such as Claude Code and Codex have completely changed my programming rhythm. They no longer just spit code for you in a dialog box, but directly reach into your directory, change files, run tests, and fix bugs.

But after having a good time, I found a huge "wall": These powerful CLI tools are ridiculously high for most people with non-technical backgrounds and who have never even touched a terminal.

The environment is not matched well, the agent is not connected, the Key is not set correctly... In order to use this productivity, an "environmental tax" has to be paid first.

So I decided to do it myself and made a little attempt called AgentHub. It is not a simple Web Shell, but the "cloud command center" that I imagine can squeeze various top Agents together to work collaboratively and allow ordinary people to use it with a click of a mouse.

---

1. Why not "use the web version directly"?

Many people will ask: OpenAI has a web version, Claude has a web version, and even various encapsulated clients. Why do you have to mess with a "terminal station"?

There are only two points at the core of my idea:

A. The qualitative change from "chat box" to "workstation"

The web version of the Agent is a "away game". You have to copy and paste the code. It can't see your entire project structure. The CLI version of Agents is a "home game" and they are in your project directory. What AgentHub needs to do is move this "home" to the cloud and give you an out-of-box workspace. You only need to upload the folder, and the remaining environment configuration, rights management, and network penetration are all automatically smoothed out in the background by the hub declaration.

B. Eliminate "isolated islands" and establish "hubs"

Agents are now isolated islands. Each top Agent has its own unique 'way of thinking' and areas of expertise. The significance of Hub is that it no longer allows you to ask multiple choice questions between different 'geniuses', but to dispatch the most suitable one at any time based on the granularity of the task.

• Now it : allows you to switch between different top-level Agents to handle the same project with one click under one interface.
• Future It : It is the most exciting part of my idea- multi-agent collaboration .

---

2. Core architecture: Not just connection, but also "choreography"

In the underlying design of AgentHub, I introduced tmux persistent sessions and dynamic reverse proxies .

Highlights of ### Concept:

1. Low Barrier : All Agents run under isolated Linux users on the host machine. You only need one account to log in, which is as simple as using flying books or Notion. The complex node-pty and tmux status behind it remains, and users are completely unaware of it.
2. Preview of "What you see is what you get": Ordinary cloud agents write code and you can't read it. I have created a dynamic preview mechanism. As long as the Agent runs a development server, the Hub will automatically generate a public preview link for you. This "real-time feedback" is extremely important for non-technical people.

---

3. Next step: From "multiple choices" to "multiple in one"

What I currently achieve is "one-stop access", but my ultimate goal is Multiple collaboration of Agents .

The scene in my imagination is: I gave an order to AgentHub: "Help me build a prototype of a backend management system with data visualization."

1. Claude Code first receives instructions, analyzes the architecture, disassembles tasks, and formulates a directory structure.
2. Codex took over and began to crazily output React component code.
3. Linter Agent starts automatically to check code quality.
4. If there is a complex logic Bug, the Debug Agent automatically intervenes to try to fix it.

Agents can even talk to each other and Review each other. This is why I insist on making a Hub - It is a container, a laboratory where the souls of different Agents can blend and work together.

---

4. Conclusion: Make AI productivity truly within reach

The birth of AgentHub actually stems from my own obsession: Good tools should not be blocked out by environment configuration.

I hope that through this attempt,"ordinary people" who are creative and thoughtful but don't understand the terminal can also have their own fully equipped AI programming workstation in the cloud.

This is just a small attempt, but I think the road to the "Agent Collaboration Network" should start with such a simple Hub.