duxweb/codux

⬛ A native connected terminal for AI agent development. 为 AI Agent 开发而生的原生互联终端。

What it solves

Codux provides a unified, high-performance workspace for developers who use multiple AI coding CLIs (such as Claude Code, Codex, and OpenCode). It eliminates the chaos of managing separate states, token costs, and remote sessions across different tools and devices, turning fragmented AI agent runs into a durable, project-aware environment.

How it works

Built with Rust and GPUI, Codux acts as a control plane rather than a text editor. It uses non-invasive wrappers and adapters to integrate with various AI CLIs, providing live status tracking and token analytics without mutating global configurations. It employs a P2P/relay system (via iroh) for end-to-end encrypted links between desktops, phones, and headless hosts, allowing users to drive long-running agent work from any device. To ensure security, it provides isolated commands (codux-ssh and codux-db) that allow agents to access servers and databases without ever seeing the actual credentials.

Who it’s for

Developers who rely on AI coding agents and CLIs for complex, multi-project work and need a way to manage long-running sessions, track token usage, and securely handle remote infrastructure access.

Highlights

  • Unified AI CLI Support: Integrates with 9+ tools including Claude Code, Codex, and Kimi Code.
  • Credential Isolation: Prevents credential leaks by injecting passwords and keys inside a helper process rather than the prompt.
  • Cross-Device Continuity: Pair phones or headless hosts to take over running sessions via encrypted P2P links.
  • Token Analytics: Detailed usage tracking by tool, model, project, and day.
  • Local-First Design: No cloud accounts required; all data, memory, and credentials stay on the user's machines.
  • WSL Integration: Native support for Windows Subsystem for Linux (WSL) runtimes.
  • Web Tunnel Browser: A proxy-isolated Chromium browser to access host-local URLs (e.g., Vite) from a controller device.

Related

  • Project
    smtg-ai/claude-squad

    A terminal application that manages multiple AI coding agents in isolated git workspaces, allowing developers to run several tasks simultaneously without conflicts.

  • Project
    anywhere-labs/Agents-Anywhere

    A remote control system that lets you manage and interact with AI coding agents (like Codex and Claude Code) running on your computer from a mobile device or web browser.

  • Project
    mensfeld/code-on-incus

    A secure sandbox for AI coding agents that uses Incus system containers to provide full machine access with active kernel-level threat detection and credential isolation.

  • Project
    milisp/codexia

    A lightweight agent workstation that provides a unified command center for Codex CLI, Claude Code, and ACP-compatible agents with a task scheduler and remote control.

  • Project
    asheshgoplani/agent-deck

    A command center for AI coding agents that allows users to manage, monitor, and orchestrate multiple agent sessions across different projects from a single terminal.