amux – terminal UI for running multiple coding agents in parallel

Introduction

The era of AI-assisted software development has fundamentally changed how developers write, review, and iterate on code. Tools like Claude Code, GitHub Copilot, Google Gemini, and OpenAI Codex are no longer experimental novelties — they are active participants in the development process. But as these agents have become more capable, a new challenge has emerged: how do you run multiple AI coding agents simultaneously across different branches, tasks, and contexts — without your terminal turning into an unmanageable mess?

Enter amux. Short for Agent Multiplexer, amux is a free, open source terminal user interface (TUI) built specifically for developers who want to coordinate multiple AI coding agents running in parallel, all from a single, organized terminal dashboard. Whether you are running Claude Code on a feature branch, Codex on a bug fix, and Gemini reviewing a refactor — amux gives you one place to see it all, switch between sessions, review diffs, and stay in control.

What is amux?

amux is a terminal-based dashboard built in Go that sits on top of tmux to orchestrate multiple AI coding agent sessions. It introduces a workspace-first model where each active task — a feature, a bug fix, an experiment — lives in its own isolated workspace, optionally backed by a Git worktree.

Unlike simply opening multiple terminal tabs or tmux panes manually, amux provides a structured interface that tracks workspace state, persists session metadata, handles agent execution, shows diff output, and allows you to jump between contexts with keyboard shortcuts or mouse clicks. It is designed for developers working with agentic coding workflows at scale — particularly those who run several long-running AI agent sessions at once and need visibility across all of them without context-switching overhead.

The Problem amux Solves

Modern AI coding agents are powerful but designed to run one session at a time. When a developer wants to work on multiple tasks simultaneously — which is the norm in professional software development — the standard approach is to open separate terminals, manage separate branches manually, and mentally track which agent is doing what in which context.

This creates several pain points that amux directly addresses:

Pain Point	Without amux	With amux
Multiple agents simultaneously	Separate terminal windows, manual tab switching	Unified TUI dashboard showing all active agent sessions
Branch isolation	Manual git checkout or stash management per task	Git worktree per workspace — branches are always isolated
Diff review	Separate git diff commands per workspace	Inline diff viewer built into the amux interface
Session persistence	tmux sessions forgotten or lost after disconnect	Persistent workspace metadata stored and reloadable
Context switching	Memorizing which terminal = which task	Named workspaces with status visible at a glance
Agent compatibility	Different agents = different launch commands to remember	Unified launcher supports Claude Code, Codex, Gemini, Amp, OpenCode, Droid

Key Features of amux

1. Parallel Agent Execution

amux is built from the ground up to run multiple coding agents simultaneously. You can have Claude Code working on a new feature in one workspace, OpenAI Codex fixing a regression in another, and Google Gemini reviewing documentation in a third — all active at the same time, all visible from one dashboard. Each agent runs inside its own isolated tmux session, so they cannot interfere with each other.

2. Workspace-First Architecture with Git Worktree Support

The central organizing concept in amux is the workspace. Every agent session is attached to a workspace, which can optionally import a Git worktree. Git worktrees allow you to check out multiple branches of the same repository simultaneously in different directories — without cloning the repo multiple times. This is a game-changer for parallel agent workflows because each agent always operates on its own branch with zero risk of merge conflicts mid-session.

# Example: Create a workspace with an isolated Git worktree
amux workspace create --name feature-auth --worktree origin/main

# List all active workspaces
amux workspace list

3. Multi-Agent Compatibility

amux supports a broad range of modern AI coding agents out of the box. The current supported agent list includes:

Agent	Provider	Primary Use Case
Claude Code	Anthropic	Full-codebase understanding, complex refactors, test generation
Codex	OpenAI	Code generation, completion, inline suggestions
Gemini	Google	Multimodal reasoning, documentation, code review
Amp	Sourcegraph	Enterprise codebase navigation and agentic coding
OpenCode	Open Source	Self-hosted, privacy-first coding agent
Droid	Community	Lightweight agent for quick scripting and automation tasks

4. Unified Interface: Agent Execution + Diff View + Terminal

One of amux’s most practical features is combining three typically separate workflows into a single TUI panel layout. Within the same interface you can monitor the active agent’s output stream in real time, view a diff of all changes the agent has made so far, and drop into a full terminal session inside that workspace. This means you never need to leave the amux dashboard to understand exactly what an agent is doing or has done.

5. Keyboard-Only and Mouse Support

amux is designed for terminal power users but accessible to those who prefer mouse interaction. Full keyboard navigation is supported with vim-style bindings for speed, while mouse clicking works for workspace switching, scrolling through output, and selecting options from menus.

6. Persistent Workspace Metadata

When you close amux or disconnect from your remote session, workspace metadata is persisted to disk. When you return, all your workspaces — their names, associated Git worktrees, agent types, and session history — are restored. The tmux session backing each workspace also persists in the background, so long-running agents are not interrupted by connection drops.

7. Built-in Operational Tooling

amux includes a suite of developer-oriented operational tools that go beyond basic agent coordination. These include structured logging for debugging agent sessions, profiling and pprof support for performance analysis of the amux process itself, debug signal dumps for diagnosing hangs or unexpected behavior, and PTY (pseudoterminal) tracing for low-level inspection of terminal I/O between amux and agent processes.

Installation

amux is available via Go install. Since it is written in Go, installation is straightforward on any system with the Go toolchain installed (Go 1.21+ recommended).Install via Go

# Install amux directly from source
go install github.com/andyrewlee/amux@latest

# Verify installation
amux --version

Prerequisites

• Go 1.21 or later installed and in PATH
• tmux 3.0+ installed (amux uses tmux for session management)
• Git 2.15+ with worktree support (for workspace isolation features)
• At least one supported AI coding agent installed (e.g., Claude Code via npm install -g @anthropic-ai/claude-code)

Platform Support

amux supports Linux and macOS. Windows is not currently supported as tmux is not natively available on Windows (though WSL2 environments work correctly).

Quick Start: Running Your First Parallel Agent Session

Here is a minimal walkthrough to get two agents running in parallel on separate branches using amux:

1. Launch amux from your project root:

d ~/projects/my-app
  amux

Create two workspaces with isolated Git worktrees:

# Workspace 1: Feature work with Claude Code
  amux workspace create --name feature-login --agent claude-code --branch feature/login

  # Workspace 2: Bug fix with Codex
  amux workspace create --name bugfix-session --agent codex --branch bugfix/session-timeout

Both agents are now running in parallel. Switch between workspaces using keyboard shortcuts (Tab, j/k) or mouse clicks.

1. Review diff output for either workspace directly in the amux interface — no need to open a separate terminal.
2. When both agents finish, review each workspace’s changes, merge the worktrees back to main, and close the workspaces:

amux workspace close --name feature-login
  amux workspace close --name bugfix-session

Conclusion

amux is a well-conceived, practically designed open source tool that addresses a real and growing pain point in modern AI-assisted software development. As coding agents become faster, smarter, and more autonomous, the bottleneck is no longer the agents themselves — it is the developer’s ability to coordinate, monitor, and manage multiple concurrent agent sessions without losing track of what is happening where.

With its workspace-first architecture, native Git worktree support, multi-agent compatibility spanning Claude Code to Gemini to Droid, built-in diff viewing, persistent metadata, and clean terminal UI, amux delivers exactly the coordination layer that parallel agentic development needs.