wangzx233/meridian

Meridian is a Chinese-language web console for managing coding tasks that run inside OpenAI's Codex CLI. Codex CLI is a command-line tool that lets a language model do work directly in a project folder. The problem Meridian tries to solve is that a developer often has Codex CLI installed on several different machines, with several different projects, and several different tasks in flight at once. Switching between them by opening new terminal windows on each machine is slow. Meridian gathers all of those machines, projects, and tasks into one browser interface so you can hop between them with a click. The project is careful to say what it is not. It does not try to replace Codex, and it does not build a new AI agent. Codex CLI still does the actual work on each target machine. Meridian's job is to run a small device agent on every machine you want to control, and then let the central console send commands to those agents, stream their output back, and remember which project and task each one belongs to. The README compares this to general-purpose agents like Hermes or OpenClaw and to IDE-based AI plugins, and positions Meridian as the lighter option focused on management rather than autonomy. The backend is written in Go, the frontend is React 19, and tasks and run events are stored in PostgreSQL 16. The recommended way to run it is Docker Compose, which starts the database, applies migrations, builds the device agent files, and serves the web interface on port 18080. A .env.example file lets the user change ports, database passwords, and bind addresses, and the README warns that public deployments should sit behind an HTTPS reverse proxy. Connecting a machine works by opening the Runner install button in the web interface, setting a Control URL the device can reach, and copying the install command for Linux, macOS, or Windows. Once installed, the device shows up in the left sidebar. A typical work session is then: create a project tied to a real working directory, create a task, send instructions turn by turn, watch streaming output in the Output, Terminal, and Files panes, and mark the task as done when finished. Notifications can be delivered through the UI, browser, or email.