Codeg

Codeg is an open-source, Apache-2.0 licensed multi-agent coding workspace from the spacering-net organization. It gives AI coding agents such as Claude Code, Codex, and Gemini CLI a single structured interface, available as a desktop app, a self-hosted server, a Docker container, and native iOS and Android clients. The latest release listed is v0.35.1.

What It Is

Codeg is an agent-agnostic workspace that talks to coding agents over the Agent Client Protocol (ACP). Fifteen agents are built in, and any other ACP-compatible agent can be added from the public registry or by pasting its distribution JSON. Every agent is rendered as the same structured conversation with tool cards, live diffs, plans, and permission prompts rather than a raw terminal.

Working with Sessions and Multiple Agents

Codeg reads the session history that each agent CLI keeps on disk, so users can import, search, and resume sessions in one place and hand them to a different agent. Typing @ delegates work to another agent, with each mentioned agent running as its own session and streaming back into the thread. To-dos can be queued to run unattended, each in its own git worktree and branch, and they wait in Review until the user accepts the diff.

The Engineering Loop

The workspace includes a file tree, editor, side-by-side diffs, a git client with a three-pane merge editor, worktrees, and an embedded terminal. Other documented pieces include:

• an infinite canvas for laying out conversations, files, and terminals
• a built-in browser
• a Computer Use preview for sharing windows with agents
• a repository panel for GitHub, GitLab, Gitea, and Forgejo issues and pull requests
• token usage reporting

Channels, Automation, and Extensions

Agents can be driven from Telegram, Lark (Feishu), and iLink (Weixin) to create tasks and approve permissions. Automations save a configured composer to run headlessly, on a cron schedule, or on demand. MCP servers and skill packs extend agents, and bundled workflows cover Office documents (.docx, .xlsx, .pptx via officecli), scientific research skills, and visual project scaffolding with Project Boot.

Deployment and Privacy

The project describes itself as local-first: parsing, storage, and project operations stay on the user's machine, and web and server modes are guarded by token authentication. The server can be installed with a script on Linux, macOS, or Windows, or run as a single Docker container. The mobile clients connect to the user's own desktop Web Service or codeg-server.