On Nov 7, 2026, a developer said on V2EX that the former codex-hub project has been upgraded to agent-hub, which manages and operates coding agents such as Codex and Claude Code in a consistent way. The project is open-sourced on GitHub at https://github.com/someonegg/agent_hub.
The tool targets the pain of running several coding agents at once: beyond switching terminals, users must find sessions, check progress, add requirements and handle confirmations. agent-hub pulls these operations into a local web workbench and also provides a CLI and skills so another AI can handle orchestration.
Router and Worker roles
The author positions agent-hub as a lightweight personal Super Agent workbench with three layers: the Router understands requirements, breaks down tasks, dispatches work and follows up on results; Workers, run as Codex, Claude Code and other sessions, carry out the execution; agent-hub provides a unified entry point for session management, interaction and reading results.
The Router can use the ahub-control skill to create or resume sessions, send tasks, read replies, view terminals, and handle interruptions and interactive confirmations. The author notes that how to split tasks and whether results meet the bar are left to the Router's judgment.
Remote dispatch from WeChat
The author recently tried a new entry point: using Tencent WorkBuddy as the Router to operate coding agents from personal WeChat. That way, when out, the user can follow progress on the computer from WeChat and use the same interface to dispatch tasks, add requirements and collect results.
The setup steps are:
- Run agent-hub on the computer and configure Codex / Claude Code;
- Install WorkBuddy and scan a QR code in the assistant settings to bind personal WeChat — no WeCom or App ID / Secret needed;
- Run ahub-cli skill path ahub-control to locate the skill directory and have WorkBuddy install ahub-control;
- In WeChat, explicitly say "use ahub-control", then describe the project and task to operate.
The author cautions that remote use requires the computer to stay powered on and connected, with WorkBuddy and agent-hub running continuously.
Author's intent and community response
My idea is simple: connect the AI tools I use every day into a personal work team. I can operate them in the workbench myself, and the Router agent can also help orchestrate.
The author invited developers who also use Codex / Claude Code to try it, and said he wants to learn how people manage multiple agents and split Router and Worker roles. As of publication, the V2EX thread had few replies and had not yet drawn large-scale discussion.
Worth watching: agent-hub's progress in supporting more coding agents, the real-world orchestration results of the ahub-control skill, and feedback on stability and security when operating remotely via WeChat.