codeg: Coordinate AI Coding Agents with Agent Client Protocol

Overview

Codeg is a multi-agent coding workspace built on the Agent Client Protocol (ACP). It brings different coding agents into a shared, structured interface so developers can review their conversations and code changes together, hand work between agents, and coordinate tasks without switching between separate agent tools.

The project targets individual developers and teams who want to use several coding agents across desktop, self-hosted server, Docker, and mobile. It also offers a command-line client for scripts and CI, with sessions and project operations described as local-first and server access protected by token authentication.

Key Features

• Connects 19 built-in coding agents and custom ACP agents through a common conversation interface.
• Imports, searches, resumes, and shares agent session history.
• Supports agent delegation and parallel work, including unattended tasks in isolated Git worktrees.
• Combines conversations with live diffs, project files, terminals, and an editor.
• Provides desktop, self-hosted server, Docker, iOS, and Android access, plus Telegram, Lark, and WeChat integrations.
• Includes a CLI for sending prompts and receiving replies or JSON events in scripts and CI.
• Adds repository workflows for issues, pull requests, checks, and merges across GitHub, GitLab, Gitea, and Forgejo.

Use Cases

• A developer comparing or assigning work to different coding agents can manage their conversations and changes in one workspace.
• A team that wants agents to work in parallel can delegate tasks and review their work in separate worktrees.
• A developer maintaining long-running agent work can search and resume session histories across supported agents.
• A remote developer can run Codeg as a self-hosted server or Docker container and connect from a browser or mobile app.
• A team can invoke an agent from scripts or CI using the CLI, for tasks such as reviewing a diff.

Getting Started

For a quick self-hosted start with Docker:

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

Open http://<host>:3080 and use the access token printed by the server. See the README and documentation for desktop installation, configuration, and other deployment options.

Alternatives

• vicoa: Vicoa focuses on self-hosted agent sessions and task management across desktop, web, mobile, and CLI, rather than an ACP-centered collaborative workspace.
• agent-orchestrator: Agent Orchestrator emphasizes Git-backed tasks in isolated worktrees, with pull requests, CI, and reviews in a desktop workspace.
• orca: Orca centers on side-by-side coding agents in isolated Git worktrees, rather than coordinating conversations and tasks across desktop, server, and mobile.
• cccc: CCCC focuses on durable agent messaging and delivery across runtimes, machines, and chat platforms, rather than a unified workspace for sessions and code changes.

Considerations

• The project is active, with frequent commits and releases and recent issue and pull-request activity, but contribution is highly concentrated in one person. That concentration is a continuity risk for a feature-rich application.
• The reported runtime requirement is Node.js. Docker and Compose are available, but the supplied project facts do not indicate a test suite.
• The README describes computer-use support as a preview, so teams should treat that capability as less mature than the main workspace features.
• The project describes itself as local-first, but server and web access require token authentication. Review the privacy and security documentation before exposing a server remotely.
• The Apache-2.0 license permits broad use, subject to its license terms and notices.