codeg

# Codeg [](https://github.com/xintaofei/codeg/releases) [](./LICENSE) [](https://tauri.app/) [](https://nextjs.org/) [](./Dockerfile) <p> <strong>English</strong> | <a href="./docs/readme/README.zh-CN.md">简体中文</a> | <a href="./docs/readme/README.zh-TW.md">繁體中文</a> | <a href="./docs/readme/README.ja.md">日本語</a> | <a href="./docs/readme/README.ko.md">한국어</a> | <a href="./docs/readme/README.es.md">Español</a> | <a href="./docs/readme/README.de.md">Deutsch</a> | <a href="./docs/readme/README.fr.md">Français</a> | <a href="./docs/readme/README.pt.md">Português</a> | <a href="./docs/readme/README.ar.md">العربية</a> </p> 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, Cursor, etc.) into one workspace, supporting conversation aggregation and multi-agent collaboration, with desktop installation plus server/Docker deployment.  ## Sponsors <table> <tr> <td align="center" width="220"> <a href="https://www.compshare.cn/?ytag=GPU_YY_git_codeg" target="_blank"><img src="./docs/images/compshare.png" alt="Compshare" width="160" /></a><br/> <strong><a href="https://www.compshare.cn/?ytag=GPU_YY_git_codeg">Compshare (UCloud)</a></strong> </td> <td>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!</td> </tr> <tr> <td align="center" width="220"> <a href="https://sui-xiang.com/register?aff=JPFCRHHBE8HE" target="_blank"><img src="./docs/images/sui-xiang.jpg" alt="随想AI中转站" width="200" /></a><br/> <strong><a href="https://sui-xiang.com/register?aff=JPFCRHHBE8HE">随想AI中转站</a></strong> </td> <td>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.</td> </tr> <tr> <td align="center" width="220"> <a href="https://hezu.ink/sign-up?aff=0wVz" target="_blank"><img src="./docs/images/hezu-ink.jpg" alt="合租巴士" width="200" /></a><br/> <strong><a href="https://hezu.ink/sign-up?aff=0wVz">合租巴士</a></strong> </td> <td>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>.</td> </tr> <tr> <td align="center" width="220"> <a href="https://onehop.ai/platform/login?ref=CODEG&utm_source=github&utm_medium=readme_sponsor&utm_campaign=codeg&utm_content=sponsor_cta" target="_blank"><img src="./docs/images/onehop.jpg" alt="OneHop" width="120" /></a><br/> <strong><a href="https://onehop.ai/platform/login?ref=CODEG&utm_source=github&utm_medium=readme_sponsor&utm_campaign=codeg&utm_content=sponsor_cta">OneHop</a></strong> </td> <td>Thanks to OneHop for sponsoring this project! OneHop gives Codeg users one OpenAI-compatible API key for hundreds of leading models, including GPT, Claude, Gemini, DeepSeek, Kimi, and Qwen. Switch models without managing multiple provider accounts or repeatedly changing your code, and pay only for what you use. <a href="https://onehop.ai/platform/login?ref=CODEG&utm_source=github&utm_medium=readme_sponsor&utm_campaign=codeg&utm_content=sponsor_cta">Sign up through Codeg</a> to receive $1 in credit, then join the OneHop community and participate in the welcome activity for an additional $5 — up to $6 in test credit in total.</td> </tr> </table> > Want to become a Codeg sponsor? [Reach out to us by email.](mailto:itpkcn@gmail.com) ## Main Interface   ## Multi-Agent Collaboration   ## Office Workflow   ## 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 - **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 - MCP management (local scan + registry search/install) - Skills management (global and project scope) - Git remote account management (GitHub and other Git servers) - Web service mode — access Codeg from any browser for remote work - **Standalone server deployment** — run `codeg-server` on any Linux/macOS server, access via browser - **Docker support** — `docker compose up` or `docker run`, with custom token, port, and volume mounts for data persistence and project directories - Runtime Logs — a live in-app log viewer with filtering and per-module log levels - Integrated engineering loop (file tree, diff, git changes, commit, terminal) ## Supported Agents | Agent | Environment Variable Path | macOS / Linux Default | Windows Default | | ------------ | ------------------------------------- | ------------------------------------- | ----------------------------------------------------- | | Claude Code | `$CLAUDE_CONFIG_DIR/projects` | `~/.claude/projects` | `%USERPROFILE%\\.claude\\projects` | | Codex CLI | `$CODEX_HOME/sessions` | `~/.codex/sessions` | `%USERPROFILE%\\.codex\\sessions` | | OpenCode | `$XDG_DATA_HOME/opencode/opencode.db` | `~/.local/share/opencode/opencode.db` | `%USERPROFILE%\\.local\\share\\opencode\\opencode.db` | | Gemini CLI | `$GEMINI_CLI_HOME/.gemini` | `~/.gemini` | `%USERPROFILE%\\.gemini` | | OpenClaw | — | `~/.openclaw/agents` | `%USERPROFILE%\\.openclaw\\agents` | | Cline | `$CLINE_DIR` | `~/.cline/data/tasks` | `%USERPROFILE%\\.cline\\data\\tasks` | | Hermes Agent | `$HERMES_HOME/state.db` | `~/.hermes/state.db` | `%USERPROFILE%\\.hermes\\state.db` | | CodeBuddy | `$CODEBUDDY_CONFIG_DIR/projects` | `~/.codebuddy/projects` | `%USERPROFILE%\\.codebuddy\\projects` | | Kimi Code | `$KIMI_CODE_HOME/sessions` | `~/.kimi-code/sessions` | `%USERPROFILE%\\.kimi-code\\sessions` | | Pi | `$PI_CODING_AGENT_SESSION_DIR` | `~/.pi/agent/sessions` | `%USERPROFILE%\\.pi\\agent\\sessions` | | Grok Build | `$GROK_HOME/sessions` | `~/.grok/sessions` | `%USERPROFILE%\\.grok\\sessions` | | Cursor | `$CURSOR_CONFIG_DIR/chats` | `~/.cursor/chats` | `%USERPROFILE%\\.cursor\\chats` | > Note: environment variables take precedence over fallback paths. <details> <summary><h2>Project Boot</h2></summary> 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. </details> <details> <summary><h2>Chat Channels</h2></summary> 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](docs/chat-channels/telegram-topic-mode.md) to bind each topic to a separate Codeg session. </details> <details> <summary><h2>Office Documents</h2></summary> 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 - **Analyze & Proofread** — inspect document structure, surface formatting issues, and proofread content - **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 </details> <details> <summary><h2>Scientific Research</h2></summary> 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 </details> <details> <summary><h2>Automations</h2></summary> 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 </details> <details> <summary><h2>Quick Start</h2></summary> ### Requirements - Node.js `>=22` (recommended) - pnpm `>=10` - Rust stable (2021 edition) - Tauri 2 build dependencies (desktop mode only) Linux (Debian/Ubuntu) example: ```bash sudo apt-get update sudo apt-get install -y \ libwebkit2gtk-4.1-dev \ libayatana-appindicator3-dev \ librsvg2-dev \ patchelf ``` ### Binaries 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 ```bash 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-<triple> # 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. #### Option 1: One-line install (Linux / macOS) ```bash curl -fsSL https://raw.githubusercontent.com/xintaofei/codeg/main/install.sh | bash ``` Install a specific version or to a custom directory: ```bash curl -fsSL https://raw.githubusercontent.com/xintaofei/codeg/main/install.sh | bash -s -- --version v0.5.2 --dir ~/.local/bin ``` Then run: ```bash codeg-server ``` #### Option 2: One-line install (Windows PowerShell) ```powershell irm https://raw.githubusercontent.com/xintaofei/codeg/main/install.ps1 | iex ``` Or install a specific version: ```powershell .\install.ps1 -Version v0.5.2 ``` #### Option 3: Download from GitHub Releases Pre-built binaries (with bundled web assets) are available on the [Releases](https://github.com/xintaofei/codeg/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` | ```bash # 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](#in-place-updates). #### Option 4: Docker ```bash # 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 ```bash 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: ```bash 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. #### Configuration Environment variables: | Variable | Default | Description | | ------------------------------ | ---------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `CODEG_PORT` | `3080` | HTTP port | | `CODEG_HOST` | `0.0.0.0` | Bind address | | `CODEG_TOKEN` | _(random)_ | Auth token (printed to stderr on start) | | `CODEG_DATA_DIR` | `~/.local/share/codeg` | SQLite database directory (also roots `uploads/`, `pets/`) | | `CODEG_STATIC_DIR` | `./web` or `./out` | Next.js static export directory | | `CODEG_MCP_BIN` | _(unset)_ | 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 `<data dir>/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". | </details> <details> <summary><h2>Architecture</h2></summary> ```text Next.js 16 (Static Export) + React 19 | | invoke() (desktop) / fetch() + WebSocket (web) v ┌─────────────────────────┐ │ Transport Abstraction │ │ (Tauri IPC or HTTP/WS) │ └─────────────────────────┘ | v ┌─── Tauri Desktop ───┐ ┌─── codeg-server ───┐ │ Tauri 2 Commands │ │ Axum HTTP + WS │ │ (window management) │ │ (standalone mode) │ └──────────┬───────────┘ └──────────┬──────────┘ └──────────┬───────────────┘ v Shared Rust Core |- AppState |- ACP Manager |- Parsers (conversation ingestion) |- Chat Channels |- Git / File Tree / Terminal |- MCP marketplace + config |- Office Tools (officecli) + Automations |- SeaORM + SQLite | ┌───────┼───────┐ v v v Local Filesystem Git Chat Channels / Git Repos Repos (Telegram, Lark, iLink) ``` </details> ## Privacy & Security - Local-first by default for parsing, storage, and project operations - Network access happens only on user-triggered actions - System proxy support for enterprise environments - Web service mode uses token-based authentication ## Community - Scan the QR code below to join our WeChat group for discussions, feedback, and updates <img src="./docs/images/weixin-light.jpg#gh-light-mode-only" alt="WeChat" width="240" /> <img src="./docs/images/weixin-dark.jpg#gh-dark-mode-only" alt="WeChat" width="240" /> - Thanks to the [LinuxDO](https://linux.do) community for their support ## Acknowledgments - [ACP](https://agentclientprotocol.com) — the Agent Client Protocol (ACP) is the foundation that enables Codeg to connect with multiple agents - [Superpowers](https://github.com/obra/superpowers) — powers Codeg's expert skills module - [OfficeCLI](https://github.com/iOfficeAI/OfficeCLI) — powers Codeg's Office documents workflow - [scientific-agent-skills](https://github.com/K-Dense-AI/scientific-agent-skills) — powers Codeg's Scientific Research skills (MIT-licensed subset) ## License Apache-2.0. See `LICENSE`.