Codeg — Summary

Codeg (Code Generation) is a multi-agent coding workspace that unifies multiple AI coding agents — Claude Code, Codex CLI, OpenCode, Gemini CLI, OpenClaw, Cline, and others — into a single desktop and server application. It aggregates conversation sessions from all supported agents into one workspace, enables multi-agent collaboration (a main agent delegates subtasks to specialized sub-agents that each run as independent sessions), and provides a visual project bootstrapping interface, built-in git worktree flows, MCP management, and skills management at both global and project scopes. Codeg ships as a Tauri 2 desktop application (Rust backend + Next.js frontend) and as a standalone server binary (codeg-server) deployable via Docker on any Linux/macOS server, accessible from any browser. Chat channel integrations (Telegram, Lark/Feishu, iLink/Weixin) allow controlling agent sessions remotely from messaging apps.

Among the seeds, Codeg most closely resembles claude-flow in providing multi-agent orchestration, but differs in that Codeg is a UI workspace rather than an SDK or command-pack: it provides a visual interface for launching and monitoring agent sessions, while claude-flow provides programmatic multi-agent APIs. Codeg is also distinguished by being the only framework in this batch that aggregates across heterogeneous agent types (not just Claude), treating agent-type diversity as a first-class feature.

Codeg — Overview

Origin

Codeg is authored by xintaofei (GitHub: xintaofei/codeg), published in 2025-2026. Written in TypeScript (Next.js frontend) + Rust (Tauri 2 backend). Reached 1,344 stars as of 2026-05-26 and is actively maintained. Sponsored by Compshare (UCloud) and MyClaw.ai.

Philosophy

The README describes the project as "a multi-agent coding workspace [that] brings multiple agents into one workspace, supporting conversation aggregation and multi-agent collaboration." The philosophy is agent-type agnosticism: rather than being opinionated about which AI coding agent is best, Codeg acts as a universal workspace that works with all of them.

Key Differentiators (from README highlights)

"Conversation Aggregation — import sessions from all supported agents into one unified workspace"

"Multi-Agent Collaboration — within a single session, the main agent delegates to sub-agents of different types (e.g. Claude Code calling Codex, Gemini) to jointly complete a task, each running as an independent session"

"Parallel development with built-in git worktree flows"

"Chat Channels — connect Telegram, Lark (Feishu), iLink (Weixin) and more to your coding agents for real-time notifications, full session interaction, and remote task control"

Target Users

• Developers who use multiple AI coding tools and want a unified workspace
• Teams wanting to delegate subtasks across agent types
• Remote developers who want to control agent sessions via mobile messaging apps

Language Support

Full internationalization (next-intl). README available in 10 languages: English, 简体中文, 繁體中文, 日本語, 한국어, Español, Deutsch, Français, Português, العربية.

Codeg — Architecture

Distribution

• Tauri 2 desktop app (macOS, Windows, Linux)
• Standalone server binary codeg-server (Linux/macOS)
• Docker: docker compose up or docker run

Required Runtime

• No runtime dependencies for end users (Tauri bundles Rust binary)
• For server mode: Docker or direct binary
• For development: pnpm + Rust + Node.js

Tech Stack

Repository Structure

codeg/
├── src/                    # Next.js frontend
├── src-tauri/              # Rust/Tauri backend
├── .cargo/                 # Rust configuration
├── public/                 # Static assets
├── CLAUDE.md               # Development guidance (Chinese)
├── docs/images/            # Screenshots
└── docker-compose.yml      # Docker deployment

Supported Agent Session Sources

Web Service Mode

The codeg-server binary runs Codeg as a web service on any Linux/macOS server. Users access it from any browser. Supports custom token, port, and volume mounts via Docker.

Codeg — Components

Core Features (UI Modules)

Multi-Agent Collaboration Architecture

Main Agent Session (e.g., Claude Code)
    │
    ├── Sub-Agent 1: Codex (implements backend task)
    ├── Sub-Agent 2: Gemini (implements frontend task)
    └── Sub-Agent 3: Claude Code (code review)

Each sub-agent runs as an independent session. The main agent coordinates via delegation.

Project Boot Options

Visual scaffolding supports:

• Frameworks: Next.js, Vite, React Router, Astro, Laravel
• Package managers: pnpm, npm, yarn, bun
• UI toolkit: shadcn/ui (with live preview of color theme, border radius, icon library, font)

Database

SQLite via SeaORM — stores session metadata, MCP registrations, and skills configurations.

Codeg — Uniqueness

Differs from Seeds

Codeg is the only framework in the corpus that provides a universal multi-agent workspace spanning heterogeneous agent types. All seeds focus on enhancing a single agent type (Claude Code primarily). Codeg treats Claude Code, Codex, Gemini CLI, OpenCode, OpenClaw, and Cline as peer workers that can be delegated to within a single workflow — a main Claude Code session can spawn a Codex sub-session and a Gemini sub-session in the same task. This cross-agent delegation pattern is absent from all 11 seeds. Claude-flow is the closest seed in spirit (hierarchical multi-agent orchestration), but claude-flow's agents are all Claude Code instances; Codeg's agents span heterogeneous types with different LLM backends. The Chat Channels integration (Telegram, Lark, iLink) for remote session control via mobile messaging is also unique in the corpus. Codeg's git worktree support resembles superpowers and claude-flow, but is surfaced in a desktop GUI rather than CLI skill invocations.

Positioning

Codeg is an "AI coding fleet manager" — a workspace that treats multiple AI agents as a workforce to be dispatched and monitored. The value proposition is: every major AI coding agent has a different strength; Codeg lets a team use all of them in one workflow. It is not opinionated about how agents should work (no spec format, no TDD enforcement, no quality gates) — it provides the environment, not the methodology.

Observable Failure Modes

1. Coordination relies on the main agent: If the main agent's delegation logic fails (e.g., incorrect prompt to spawn sub-agents), sub-agent work is disconnected from the goal.
2. No shared context between agents: Sub-agents in different session types (Codex, Gemini) don't natively share Claude Code's context or memory. Coordination is coarse-grained.
3. Agent-type version drift: When Claude Code, Codex, or Gemini update their session formats, Codeg's session reader may break.
4. No conflict detection: Unlike Asynkor, Codeg has no mechanism to prevent two agents from editing the same file simultaneously in a multi-agent session.