Skip to content

Getting Started with AgentMux

AgentMux is an agentic workflow environment — a desktop app for running, observing, and coordinating AI agents. Built on Rust with a bundled Chromium 146 via CEF and a SolidJS frontend.

AgentMux gives you a multiplexed workspace where multiple AI agents run in parallel panes. Seven providers are supported as first-class agent types: Claude Code, Codex CLI, Gemini CLI, OpenClaw, Kimi Code CLI, GitHub Copilot CLI, and Pi. You see every tool call, file write, and network request in real time. Panes communicate with each other through a backend WebSocket pub-sub event system.

  • Panes — Individual workspace units. Each pane has a type (agent, browser, terminal, editor, swarm, drone, sysinfo, help).
  • Identity bundles — Named credential sets (GitHub PAT, AWS profile, Anthropic API key, …) you assign to an agent at launch. Survives renames; swappable without restart.
  • Memory bundles — Reusable agent personality + capability stacks (provider, model, instructions, MCP, skills). Manage via the Memory tab inside an Agent pane’s settings panel. Replaces the older “Forge” concept.
  • Interagent Comms — Panes communicate via a backend pub-sub event system. An agent’s output can stream into another pane’s input.
  • Subagent monitoring — The Swarm pane provides a bird’s-eye view of all sub-agents spawned by primary agents; clicking one opens a focused Subagent view.
  • Reducer stack — A 4-layer audited dispatch model (launcher / host / sidecar / frontend slices). Every state mutation is structured and logged.

Download the .dmg from agentmux.ai or the GitHub releases.

Download the installer (.exe) or portable (.zip) from agentmux.ai.

Download the AppImage from agentmux.ai:

Terminal window
chmod +x AgentMux_amd64.AppImage
./AgentMux_amd64.AppImage