Primary Purpose:
Codeg is a multi-agent coding workspace designed to unify local AI coding agents such as Claude Code, Codex CLI, OpenCode, Gemini CLI, and others into a single platform. This tool enables efficient development by integrating various AI-driven tools seamlessly.
Key Features:
Multi-agent collaboration in one workspace.
Remote browser access for development from any location.
Conversation aggregation for structured interactions.
Parallel git worktree development enhancing productivity.
Comprehensive MCP/Skills management capabilities.
Integration with chat channels like Telegram and Lark.
Integrated Git/file/terminal workflows for streamlined operations.
Audience & Benefit:
Ideal for developers, DevOps teams, and enterprises utilizing AI-driven coding tools, Codeg streamlines workflow by integrating multiple agents into a single platform. This integration saves time, enhances efficiency, and facilitates collaboration across various AI tools, making it a valuable asset in modern development environments.
This description maintains a professional tone, focuses on benefits without hype, and adheres to the specified structure and guidelines.
Codeg (Code Generation) is a multi-agent coding workspace. It brings multiple agents (Claude Code, Codex CLI, OpenCode, Gemini CLI, OpenClaw, Cline, Hermes Agent, CodeBuddy, Kimi Code, Pi, Grok Build, etc.) into one workspace, supporting conversation aggregation and multi-agent collaboration, with desktop installation plus server/Docker deployment.
Sponsors
<a href="https://www.compshare.cn/?ytag=GPU_YY_git_codeg"><img src="./docs/images/compshare.png" width="160" /></a>
<strong><a href="https://www.compshare.cn/?ytag=GPU_YY_git_codeg">Compshare (UCloud)</a></strong>
Thanks to Compshare for sponsoring this project! Compshare is UCloud's AI cloud platform, offering cost-effective monthly and pay-as-you-go agent Plan subscriptions for Chinese models, starting at just ¥49/month. It also provides stable officially-proxied access to overseas models. Supports Claude Code, Codex, and API integrations. Enterprise-ready with high concurrency, 24/7 technical support, and self-service invoicing. Users who sign up via <a href="https://www.compshare.cn/?ytag=GPU_YY_git_codeg">this link</a> receive ¥5 in free platform credits!
<a href="https://sui-xiang.com/register?aff=JPFCRHHBE8HE"><img src="./docs/images/sui-xiang.jpg" width="200" /></a>
<strong><a href="https://sui-xiang.com/register?aff=JPFCRHHBE8HE">随想AI中转站</a></strong>
Thanks to 随想AI中转站 for sponsoring this project! 随想AI中转站 is a reliable and efficient API relay provider, offering relay services for Claude, Codex, Gemini, and more. New accounts earn ¥0.5 in test credit with each daily check-in after <a href="https://sui-xiang.com/register?aff=JPFCRHHBE8HE">signing up</a>; top-ups are credited 1:1 — no subscription, pay as you go. Multi-route redundancy, cross-region disaster recovery, and automatic failover keep long-lived SSE connections uninterrupted.
<a href="https://hezu.ink/sign-up?aff=0wVz"><img src="./docs/images/hezu-ink.jpg" width="200" /></a>
<strong><a href="https://hezu.ink/sign-up?aff=0wVz">合租巴士</a></strong>
Thanks to 合租巴士 for sponsoring this project! 合租巴士 is a reliable and efficient AI relay platform, offering highly stable relay for mainstream models such as Codex and Claude Code. Top-ups are credited at a transparent 1:1 ratio, with Codex rate subsidies as low as 0.08. <a href="https://hezu.ink/sign-up?aff=0wVz">Join the group via the official website to get $5 in trial credit</a>.
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
Project Boot — visually scaffold new projects with live preview
Office Documents — create, analyze, proofread, and edit .docx / .xlsx / .pptx through the bundled officecli toolset, with live in-tab preview that refreshes as the agent edits
Scientific Research — bundled science skills (hypothesis generation, experimental design, statistics, visualization, critical appraisal, literature search) any agent can invoke, managed per-agent
Automations — save a composer setup as a reusable automation that runs headlessly, on a cron schedule or on demand
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
> Note: environment variables take precedence over fallback paths.
Project Boot
Create new projects visually with a split-pane interface: configure on the left, preview in real time on the right.
What it does
Visual Configuration — pick style, color theme, icon library, font, border radius, and more from dropdowns; the preview iframe updates instantly
Live Preview — see your chosen look & feel rendered in real time before creating anything
One-Click Scaffolding — hit "Create Project" and the launcher runs shadcn init with your preset, framework template (Next.js / Vite / React Router / Astro / Laravel), and package manager of choice (pnpm / npm / yarn / bun)
Package Manager Detection — automatically checks which package managers are installed and shows their versions
Seamless Integration — the newly created project opens in Codeg's workspace right away
Currently supports shadcn/ui project scaffolding, with a tab-based design ready for more project types in the future.
Chat Channels
Connect your favorite messaging apps — Telegram, Lark (Feishu), iLink (Weixin), and more — to your AI coding agents. Create tasks, send follow-up messages, approve permissions, resume sessions, and monitor activity — all from your chat app. Receive real-time agent responses with tool-call details, permission prompts, and completion summaries without ever opening a browser.
Supported Channels
Channel
Protocol
Status
Telegram
Bot API (HTTP long-polling)
Built-in
Lark (Feishu)
WebSocket + REST API
Built-in
iLink (Weixin)
WebSocket + REST API
Built-in
> More channels (Discord, Slack, DingTalk, etc.) are planned for future releases.
Telegram forum supergroups can also use Telegram topic mode to bind each topic to a separate Codeg session.
Office Documents
Work with Word, Excel, and PowerPoint files as a first-class workflow. The bundled officecli toolset lets your agents create, analyze, proofread, and edit .docx, .xlsx, and .pptx documents — and you can preview the result right inside Codeg.
What it does
Create & Edit — generate new documents or modify existing .docx / .xlsx / .pptx files, including charts, tables, and formatting
Live Preview — open a .docx / .xlsx / .pptx in a file tab and it renders inline, refreshing automatically as the agent edits — backed by a long-lived officecli watch server (reverse-proxied and capability-authenticated so it works in web and standalone-server deployments)
Quick Actions — the welcome page offers Coding, Office, and Scientific Research tabs that drop the matching skill invocation and a prompt template into the composer with one click; a skill that isn't enabled for the selected agent shows a lock badge linking to where you can turn it on
Office Tools settings — a dedicated settings page installs officecli and manages its document skills through a skill-by-agent matrix: toggle any (skill, agent) pair, flip a skill across all agents or every skill for one agent, and apply bulk changes at once
Scientific Research
Turn any agent into a rigorous research assistant. Codeg bundles a curated set of MIT-licensed scientific-research skills — from ideation to analysis to write-up — that install into the shared central skill store and link into whichever agents you choose, exactly like the expert and office toolsets.
What it does
Curated skills — hypothesis generation, experimental design, statistical power, statistical analysis, exploratory data analysis, scientific visualization, critical appraisal, peer review, citation management, scholar evaluation, paper lookup, and AI schematics
Quick Actions — the welcome page's Scientific Research tab drops the matching skill invocation plus a localized prompt template into the composer with one click
Science settings — a dedicated settings page manages the skills through a skill-by-agent matrix, with badges flagging skills that need an API key or a Python environment
Automations
Turn any composer setup — agent, model, prompt, working directory, and options — into a reusable Automation that runs without opening the UI.
What it does
Save once, reuse — capture a fully-configured composer as a named, reusable automation
Scheduled or on demand — run it on a cron schedule or trigger it manually whenever you need it
Headless execution — automations run in the background and create real sessions you can open in the workspace at any time, then return you straight to the workspace when you start one
Codeg ships three Rust binaries from a single workspace:
Binary
Role
Build
codeg
Tauri desktop app (window, tray, updater)
pnpm tauri build (release) / pnpm tauri dev (dev)
codeg-server
Standalone HTTP + WebSocket server for browser/headless deployments
pnpm server:build / pnpm server:dev
codeg-mcp
Per-launch stdio MCP companion that surfaces the delegate_to_agent tool to agent CLIs (multi-agent collab)
pnpm tauri:prepare-sidecars (auto-invoked by tauri dev / tauri build)
codeg-mcp must sit next to its parent binary at runtime — installers, the Docker image, and the Tauri sidecar bundler all place it next to codeg / codeg-server. Source builds and custom layouts can override the lookup with the CODEG_MCP_BIN=/abs/path/codeg-mcp env var. If the companion is missing, delegation is skipped (a single warning is logged) and the rest of the agent session keeps working.
Development
pnpm install
# Frontend only (Next.js dev server, no Rust)
pnpm dev
# Frontend static export to out/
pnpm build
# Full desktop app (Tauri + Next.js, builds codeg-mcp sidecar automatically)
pnpm tauri dev
# Desktop release build (bundles codeg-mcp as externalBin)
pnpm tauri build
# Standalone server (no Tauri/GUI required)
pnpm server:dev
pnpm server:build # release binary at src-tauri/target/release/codeg-server
# Build the codeg-mcp companion explicitly (for the host triple)
pnpm tauri:prepare-sidecars # output: src-tauri/binaries/codeg-mcp-
# Skip sidecar prep when iterating on the frontend and you don't need delegation
CODEG_SKIP_SIDECAR=1 pnpm tauri dev
# Lint
pnpm eslint .
# Frontend tests (vitest)
pnpm test
pnpm test:watch
pnpm test:coverage
# Rust checks (run in src-tauri/)
cargo check # desktop (default features)
cargo check --no-default-features --bin codeg-server # server mode
cargo check --no-default-features --bin codeg-mcp # MCP companion
cargo clippy --all-targets --features test-utils -- -D warnings
# Rust tests
cargo test --features test-utils # desktop (incl. integration)
cargo test --no-default-features --bin codeg-server --lib # server mode
cargo insta review # accept parser snapshot updates
> Tip: when you have a fresh codeg-mcp build under src-tauri/target/release/ and want to point a manually-launched codeg-server at it without reinstalling, export CODEG_MCP_BIN=$(pwd)/src-tauri/target/release/codeg-mcp.
Server Deployment
Codeg can run as a standalone web server without a desktop environment.
Pre-built binaries (with bundled web assets) are available on the Releases page:
Platform
File
Linux x64
codeg-server-linux-x64.tar.gz
Linux arm64
codeg-server-linux-arm64.tar.gz
macOS x64
codeg-server-darwin-x64.tar.gz
macOS arm64
codeg-server-darwin-arm64.tar.gz
Windows x64
codeg-server-windows-x64.zip
# Example: download, extract, and run
tar xzf codeg-server-linux-x64.tar.gz
cd codeg-server-linux-x64
CODEG_STATIC_DIR=./web ./codeg-server
> For unattended deployments, start it with --supervise so a failed in-place upgrade is automatically rolled back — see In-place updates.
Option 4: Docker
# Using Docker Compose (recommended)
docker compose up -d
# Or run directly with Docker
docker run -d -p 3080:3080 -v codeg-data:/data ghcr.io/xintaofei/codeg:latest
# With custom token and project directory mounted
docker run -d -p 3080:3080 \
-v codeg-data:/data \
-v /path/to/projects:/projects \
-e CODEG_TOKEN=your-secret-token \
ghcr.io/xintaofei/codeg:latest
The Docker image uses a multi-stage build (Node.js + Rust → slim Debian runtime) and includes git and ssh for repository operations. Data is persisted in the /data volume. You can optionally mount project directories to access local repos from within the container.
Option 5: Build from source
pnpm install && pnpm build # build frontend
cd src-tauri
cargo build --release --bin codeg-server --no-default-features
cargo build --release --bin codeg-mcp --no-default-features # delegation companion
CODEG_STATIC_DIR=../out ./target/release/codeg-server # codeg-mcp is picked up as a sibling
If you keep the two binaries in separate directories, set CODEG_MCP_BIN=/abs/path/to/codeg-mcp so the runtime can still find the companion; without it, multi-agent delegation is silently disabled.
In-place updates
The server can update itself from Settings → Software Update: it downloads the signed release for its platform, swaps the binaries and web assets on disk, and restarts — no manual re-deploy. This is Linux/macOS only (disabled on Windows). The previous version is kept as a backup, so the same screen offers a Roll back action to return to it.
Run under the supervisor for auto-rollback. Start the standalone server with --supervise so a freshly-upgraded process that fails to boot within the trial window is automatically reverted to the previous version:
CODEG_STATIC_DIR=./web ./codeg-server --supervise
Without --supervise the server still updates in place (it re-execs itself), but the upgrade is best-effort: there is no supervisor to auto-roll-back a version that can't start. The Docker image already runs under the supervisor.
Docker upgrades change the container, not the image. An in-place upgrade rewrites the binaries and web assets inside the running container's writable layer, so they live only in that container. The /data volume persists, but the upgraded files do not: recreating the container — docker compose up --force-recreate, a fresh docker run, or recreating after a docker pull — starts from the image again and drops the in-place upgrade. (A docker pull on its own only refreshes the local image; nothing reverts until the container is recreated.) To make an upgrade permanent, build or pull an image at the new version and recreate the container from it.
Absolute path to the codeg-mcp companion. Overrides the default sibling-of-executable + PATH lookup. Use this for source builds or custom layouts where the companion lives outside the server's install directory.
CODEG_SKIP_SIDECAR
(unset)
Frontend-only convenience for pnpm tauri dev / pnpm tauri build — when 1, skips building the codeg-mcp sidecar. Delegation is disabled in that build; ship-quality artifacts must leave it unset.
CODEG_UPLOAD_MAX_TOTAL_BYTES
(unset)
Hard cap on total bytes resident under /uploads/. Plain decimal byte count (e.g. 10737418240 for 10 GiB). Unset, 0, or an unparseable value disables the cap and prints a startup line so the posture is visible. The cap is enforced within a single codeg-server process — horizontally-scaled deployments sharing one uploads/ volume need external coordination (file lock, Redis, reverse-proxy quota).
CODEG_UPLOAD_QUOTA_STRICT
(unset)
When truthy (1 / true / yes / on), abort startup with exit code 2 if CODEG_UPLOAD_MAX_TOTAL_BYTES is set to an unparseable value, instead of fail-open with a WARN. Use this when your security policy requires "configured quota must be effective".
