We're entering an era where every employee and department can use AI to build their own applications. This creates a new challenge: when these applications need to run on the public internet with proper authentication, maintaining separate login and permission systems for each one becomes impossibly inefficient. Over the past few days I built Central — a hub system — to solve exactly this problem. Its foundation is built entirely with an agent, and it serves as shared infrastructure for everyone.
Core Services
The hub's most basic service is user management. It pulls the company's complete contact list and organizational structure directly from WeCom, then manages it automatically — when employees join or leave, the hub syncs and updates in real time. The hub also functions as a gateway. When a user's client sends a request to reach an application, it passes through the hub's gateway first. The gateway authenticates based on the cookie or header, determines the username, writes it into the request header, and forwards the request to the correct application. This architecture lets every department and employee focus entirely on implementing their application's business logic, without worrying about token management or infrastructure.
Developer Experience
Beyond authentication, the hub provides automatic deployment, continuous integration and delivery, and Git-based version management — all already implemented. The core purpose is to simplify AI-assisted software development by non-professional developers. These users only need to send one document from the website — the guide — to their AI assistant. Their AI will understand the hub's architectural logic and how applications should be implemented within it, then automatically build the required functions and adjust based on user feedback.
Shared Capabilities
The hub also provides function and service interfaces to applications. I've completed a large model management module: within the hub you can configure different AI models like DeepSeek and Alibaba Cloud's Qwen, and applications can call this interface to add AI capabilities. There's a notification function that integrates WeCom and email logic, so applications can send notifications by calling the hub's interface directly.
When an application deploys automatically, there's a preview and release process that includes database backup. File management is handled through configured Alibaba Cloud and Cloudflare object storage, letting applications save user-submitted files on domestic or overseas servers as needed. Automatic code and database backup enables one-click rollback to past versions. And because the AI model interface returns token consumption data, we can track exactly how much each department's AI usage actually costs.
Current Status
I finished testing today and successfully deployed a Golang system through the hub. This work feels meaningful, and I hope it provides some useful ideas for others thinking about infrastructure in this new era.