Muxel

A native, GPU-accelerated multi-agent terminal multiplexer for running many
AI coding agents (Claude, opencode, Amp, …) and shells side by side — built in
Rust on GPUI (Zed's UI framework).

muxel is shaped like an agent manager, not "another terminal app": a tiled,
tabbed workspace where each pane is a real terminal running a coding agent or
shell, with first-class git worktrees, live agent-status tracking, scheduled
runs, and notifications. It's a native desktop GUI (not a TUI) — every pane
embeds a real terminal emulator over a PTY.

Features

- Tiled, tabbed panes — split horizontally/vertically, drag tabs and panes to
  re-dock (Zed-style), pop a pane out into its own window, maximize, and
  double-click a divider to even out a split.
- Agents & presets — each pane launches from a preset (program, args, model,
  system prompt, env, status markers). Built-ins for Claude, opencode, Amp and
  more, all editable in Settings.
- Live agent status — each pane shows running / working / awaiting-input /
  done, detected from the terminal output, with desktop + in-app notifications on
  transitions and a cross-project dashboard.
- Git worktrees — first-class, named, color-coded worktrees shared by panes,
  with a review/dispose flow and GitHub PR helpers (via the gh CLI).
- Runners & Loops — save a prompt + agent as a *runner* to launch on demand,
  or a *Loop* to run it on a schedule (every N minutes/hours, or a daily time).
- Shared memory — opt a project into a .muxel/MEMORY.md that agents read
  and append lessons to across runs.
- Remote development (SSH) — projects that live on a remote host: shells and
  agents run there over one multiplexed SSH connection, with remote git, a remote
  file browser, persistent tmux sessions, and roaming layouts (your pane
  layout…