A Rust-native workspace for AI coding CLIs.
Run Codex, Claude Code, Gemini CLI, OpenCode, Kiro CLI, CodeWhale, and Agy across projects, worktrees, terminals, Git, memory, tokens, SSH, and mobile handoff.
Website · Download · Mobile · WeChat · Feedback
English | 简体中文
AI coding CLIs are powerful, but serious work quickly spreads across projects, Git worktrees, terminals, sessions, tokens, remote shells, and half-remembered context. Codux AI turns that scattered workflow into one durable desktop workspace.
| When AI coding gets messy | Codux AI gives you |
|---|---|
| Every AI CLI has its own state | One project-aware runtime view for Codex, Claude Code, Gemini CLI, OpenCode, Kiro CLI, CodeWhale, and Agy. |
| Long sessions are hard to resume | Live AI runtime status, local history indexing, session restore, and per-worktree context. |
| Parallel tasks collide | A worktree-first task model where each task keeps its own terminal layout, Git state, files, and AI sessions. |
| Token usage is vague | Usage by tool, model, project, worktree, and day, without maintaining spreadsheets. |
| Project context gets lost | Local memory for user habits, project profiles, module notes, and app-managed context injection. |
| Server access is fragile | Saved SSH profiles, connection testing, and a codux-ssh command that AI tools can use without seeing credentials. |
| You leave the desk mid-run | Codux Mobile pairs with the desktop host through the v3 relay/WebRTC path so you can continue sessions remotely. |
Codux AI is not an editor replacement. It is a control plane for developers who already use AI coding CLIs heavily and need a stable way to manage multi-project, long-running AI work.
Codux models real AI work as Project -> Worktree / Task -> Terminals, Files, Git, AI Sessions.
- Create Git worktrees for parallel tasks without mixing branch state.
- Switch tasks while preserving terminal tabs, splits, bottom-panel height, active AI sessions, file context, and Git state.
- Review worktree changes, compare against the base branch, merge back to the mainline, and clean up finished worktrees.
- Keep AI history and runtime activity tied to the current worktree while project-level memory remains shared.
This is the main difference from a plain terminal multiplexer: Codux knows which project and worktree each terminal belongs to, then restores the workspace around that relationship.
Codux detects supported CLIs from managed terminals, reads local session history where available, and installs app-managed hooks or memory files for tools that support them.
| Tool | Runtime Status | History Index | Resume | Memory Injection |
|---|---|---|---|---|
| Codex | Full | Full | Full | Yes |
| Claude Code | Full | Full | Full | Yes |
| Gemini CLI | Full | Full | Tool-dependent | Yes |
| OpenCode | Full | Full | Tool-dependent | Yes |
| Kiro CLI | Full | Full | Tool-dependent | Yes |
| CodeWhale | Full | Full | Tool-dependent | Yes |
| Agy | Full | Full | Tool-dependent | Yes |
Full means Codux can track that capability from the normal terminal workflow. Tool-dependent means Codux can preserve the workspace and history, while the exact resume behavior still depends on the CLI itself.
Codux is not just wrapping a shell. Each supported AI tool is represented by a runtime driver that keeps the integration path consistent:
- Hooks capture starts, completions, interruptions, permission waits, and model/session metadata.
- Probes detect current running sessions, tools, models, and accumulated usage.
- History sources normalize local CLI transcripts into one timeline.
- Memory injection gives supported CLIs project context without duplicating wrapper logic.
This keeps multiple Codex, Claude, CodeWhale, or other sessions from crossing state, and makes new tool support easier to add without rewriting the whole runtime.
Codux indexes AI session history locally and turns it into usable project context.
- See recent sessions by project and worktree.
- Track token usage by day, model, tool, project, and workspace.
- Extract user preferences, project profiles, and module notes into local memory.
- Queue memory extraction in the runtime so the UI stays responsive.
- Inject relevant context back into supported AI CLIs when launching them.
Memory and history are stored locally. Codux treats project lists and memory as the durable assets; AI history can be rebuilt from supported local CLI transcripts.
Codux keeps the terminal next to the project surfaces you need during AI work:
- Browse files, preview assets, and drag file paths into the terminal.
- Review Git changes, stage diffs, inspect history, pull, push, and handle worktree merges.
- Save SSH profiles with password or private-key credentials.
- Test SSH connectivity before saving.
- Connect from the SSH panel or let AI CLIs use the injected
codux-ssh <profile>command.
codux-ssh references a saved profile by id. It does not expose saved passwords, key passphrases, or raw connection details to the AI CLI prompt.
Database and other secure connection profiles are planned. Today, database access should be handled through your existing CLI tools, SSH tunnels, or remote shell workflow.
Codux Mobile connects to the desktop host through the v3 remote path.
- Pair mobile with the desktop using a short-lived QR ticket.
- Use the global public relay by leaving the relay setting empty, choose the China node when needed, or configure a custom relay endpoint.
- Prefer WebRTC DataChannel when a direct path is available and fall back to WebSocket relay when P2P cannot connect.
- Keep projects, terminals, files, and AI sessions running on the desktop host while mobile controls the session remotely.
Terminal input, output, file payloads, project lists, and AI stats are encrypted between Codux Desktop and Codux Mobile.
Codux includes optional desktop companions that grow with your AI coding habits. Pets can react to usage, reminders, and AI work patterns, and you can import Codex-style custom pet packages from Petdex using a flat pet.json + spritesheet.png format.
- Download Codux from GitHub Releases or codux.dux.cn.
- Install it:
- macOS: open the
.dmgand drag Codux to Applications. - Windows: run the
setup.exeinstaller.
- macOS: open the
- Open a project folder.
- Start an AI CLI in the integrated terminal.
- Optional: create a worktree task, connect an SSH profile, or pair Codux Mobile.
Recommended downloads:
| Platform | File |
|---|---|
| macOS | codux-*-macos-*.dmg |
| Windows | codux-*-windows-x86_64-setup.exe |
Updater archives and latest.json are published for automatic updates, fallback testing, and automation. Most users should download one of the two installers above.
| Action | Shortcut |
|---|---|
| New Split | ⌘T |
| New Tab | ⌘D |
| Toggle Git Panel | ⌘G |
| Toggle AI Panel | ⌘Y |
| Switch Project | ⌘1 - ⌘9 |
All shortcuts can be customized in Settings > Shortcuts.
GitHub README does not render third-party iframe players. Watch the demo on Bilibili.
Scan the QR code to add the author on WeChat and ask to join the DUXAI community group.
cargo runUseful checks before submitting changes:
cargo check
cargo test -p codux-runtime ssh::tests
node scripts/release/test-package-gpui.mjsDesktop releases are created by pushing a version tag such as v1.6.2. The release workflow builds Rust-native macOS and Windows artifacts, publishes the GitHub Release, and updates the configured updater channel.
- macOS 14.0 (Sonoma) or later
- Windows 11
Found a bug or have a feature request? Open an issue on GitHub.
For bug reports, use Help -> Export Diagnostics and attach the generated .zip. It includes runtime logs, rotated logs, performance summaries, saved app state, invalid state backups, and matching macOS diagnostic reports when available.
Manual log paths:
~/Library/Application Support/Codux/logs/runtime-rust.log~/Library/Application Support/Codux/logs/performance-summary.json%APPDATA%\Codux\logs\runtime-rust.log
Thanks to everyone who has contributed code, issues, testing, and feedback to Codux.
Wanted to be dmux, but that name was taken. So it's Codux now, which sounds like "Cool Dux" in Chinese.