Codeg: Collaborative Multi-Agent AI Coding Workspace for Developers

Introduction

Codeg (Code Generation) is a powerful multi-agent AI coding workspace that consolidates the capabilities of numerous AI coding agents into a single, cohesive platform. Developed in Rust, Codeg aims to streamline the development workflow by allowing agents to work together on tasks, manage code, and even generate office documents, all within an integrated environment. It supports a wide array of agents, including Claude Code, Codex, OpenCode, Pi, and Grok Build, providing a unified space for all your AI-assisted coding needs.

Why Use Codeg and Key Benefits

Codeg offers a comprehensive suite of features designed to boost productivity and collaboration in AI-driven development:

• Multi-Agent Collaboration: Easily delegate tasks to different AI agents using @-mentions. Agents can run in parallel, tackling complex problems together, with their work streamed directly into your conversation thread. This eliminates context switching and copy-pasting between tools.
• Task Management with To-dos: For tasks that don't require immediate oversight, Codeg's To-dos board allows agents to work on separate Git worktrees. Each task runs on its own branch, awaiting your review before changes are merged, ensuring isolated and controlled development.
• Unified Workspace: The platform provides a complete engineering loop, including a file tree, editor with live diffs, a full Git client, and an embedded terminal. You can link multiple directories to create a single, expansive workspace.
• Office Document Generation: Agents can create, analyze, and edit .pptx, .docx, and .xlsx files with a live in-tab preview. Edits are reflected instantly, allowing for iterative document creation without leaving Codeg.
• Cross-Platform Deployment: Codeg is highly flexible, deployable as a native desktop application for macOS, Windows, and Linux, a standalone codeg-server accessible via any browser, or as a Docker container. This ensures you can work in your preferred environment.
• Mobile Accessibility: Native iOS and Android clients connect to your desktop app's Web Service or codeg-server, allowing you to start sessions, monitor replies, and approve permissions from your mobile device, keeping you connected to your work from anywhere.
• Conversation Aggregation: Import and resume past sessions from all supported agents, creating a searchable history. Agents can reference previous conversations, even if initiated by a different agent, ensuring continuity.
• Custom Agents: Beyond the fifteen built-in agents, Codeg allows you to register any other ACP-compatible agent, integrating it seamlessly into your workflow.
• Split View: Organize your workspace by splitting the conversation area into multiple tab groups, enabling side-by-side work on different tasks or with different agents.

Installation

Codeg provides multiple installation options to suit your workflow:

Desktop Application

Download the installer for macOS, Windows, or Linux from the Codeg Releases page. Follow the detailed installation instructions provided in the documentation.

Server (Headless)

Run Codeg as a headless server, accessible from any browser. On Linux or macOS, use:

curl -fsSL https://raw.githubusercontent.com/xintaofei/codeg/main/install.sh | bash
CODEG_STATIC_DIR=/usr/local/share/codeg/web codeg-server

On Windows, in PowerShell:

irm https://raw.githubusercontent.com/xintaofei/codeg/main/install.ps1 | iex
$env:CODEG_STATIC_DIR="$env:LOCALAPPDATA\codeg\web"; codeg-server

Docker

Deploy the server in a single Docker container:

docker run -d -p 3080:3080 -v codeg-data:/data ghcr.io/xintaofei/codeg:latest

Mobile Clients

Install the native iOS app or the Android APK. Connect it to your desktop app's Web Service or your codeg-server by providing the URL and token. Pairing steps are detailed in the Mobile apps documentation.

Examples

Imagine you need to refactor a complex function and simultaneously review a pull request. With Codeg, you can open two split views. In one pane, you might instruct Claude Code to refactor the function. In the other, you could ask Codex to review a diff from a pull request. Both agents work in parallel, streaming their progress directly into their respective conversation threads, allowing you to monitor and interact with both tasks without switching applications.

For larger, asynchronous tasks, the To-dos board is invaluable. You can create a task, assign an agent, and Codeg will automatically set up a Git worktree on a new branch for that task. The agent works autonomously, and once complete, the task moves to a review column. You can then review the changes, request further iterations, or accept them, allowing Codeg to handle the merge process securely.

Links

• GitHub Repository: https://github.com/xintaofei/codeg
• Official Documentation: https://docs.codeg.app
• iOS App: https://apps.apple.com/app/codeg-client/id6785199071
• Android App: https://github.com/xintaofei/codeg-android/releases/latest