An enterprise-grade multi-agent coding workspace that unifies session management, parallel development, MCP/Skills orchestration, and chat channel interactions for local AI coding agents.
Codeg is an enterprise-grade multi-agent coding workspace, open-sourced under Apache-2.0 by author xintaofei. The project addresses the fragmentation of multiple local AI coding agents (Claude Code, Codex CLI, OpenCode, Gemini CLI, Cline, etc.) by providing unified session aggregation and structured rendering of conversation histories.
Architecture Overview#
Codeg uses Next.js 16 (static export) + React 19 for the frontend, Tauri 2 as the desktop framework, and a shared Rust core layer (with Axum HTTP + WebSocket server) for the backend, backed by SQLite + SeaORM. A transport abstraction layer shields the frontend from the differences between Tauri IPC and HTTP/WebSocket, enabling the same frontend codebase to serve both desktop application and standalone web server deployment modes, with Docker containerization as a third option.
Three deployment modes: desktop app, standalone server, Docker container. Supports macOS, Linux, and Windows.
Core Capabilities#
- Multi-agent unified management: Connects to multiple local coding agents via ACP (Agent Client Protocol), ingesting and parsing local session logs with structured conversation rendering
- Parallel development: Built-in git worktree workflow for driving different agents on multiple branches simultaneously
- Project Boot: Visual scaffolding tool with split-pane configuration + live preview, supporting shadcn/ui templates across Next.js / Vite / React Router / Astro / Laravel
- MCP & Skills management: Local scanning, registry search, and installation for MCP tools; global and project-scoped Skills management
- Integrated engineering loop: File tree browsing, diff viewing, git changes and commits, built-in terminal
Chat Channels#
Chat Channels is a key differentiator, currently supporting Telegram, Lark (Feishu), and iLink (WeCom):
- Initiate/continue/cancel agent tasks from chat apps
- Agent permission approval (
/approve,/deny) - Real-time event notifications (tool calls, errors, completion summaries)
- Query commands (
/search,/today,/status) - Scheduled daily reports (conversation stats, agent type distribution, project activity)
- Message templates in 10 languages
Discord, Slack, and DingTalk are marked as planned with no confirmed timeline.
Privacy & Security#
Codeg follows a local-first approach where session parsing and storage happen entirely locally. It uses token-based authentication, stores credentials in the OS Keyring, and supports enterprise system proxies.
Installation#
One-line script:
curl -fsSL https://raw.githubusercontent.com/xintaofei/codeg/main/install.sh | bash
codeg-server
Docker:
docker run -d -p 3080:3080 -v codeg-data:/data ghcr.io/xintaofei/codeg:latest
Pre-built binaries are available on GitHub Releases for Linux x64/arm64, macOS x64/arm64, and Windows x64.
Environment variables:
| Variable | Default | Description |
|---|---|---|
CODEG_PORT | 3080 | HTTP port |
CODEG_HOST | 0.0.0.0 | Bind address |
CODEG_TOKEN | Random | Auth token |
CODEG_DATA_DIR | ~/.local/share/codeg | Database directory |
Unconfirmed Information#
- ACP (Agent Client Protocol) protocol details: mentioned in README but no external spec link provided; whether it's a custom protocol is unconfirmed
- OpenClaw agent's ecosystem maturity and actual integration depth are unconfirmed
- No standalone documentation site; all docs are embedded in the GitHub repository
The project is at version v0.9.5 with 704 commits and 70 releases, indicating active iteration.