Why amux

Claude Code at scale is painful

Running more than one agent in parallel exposes problems that don't exist in a single terminal session.

Crashes at 3am

Context overflow kills sessions overnight. You wake up to nothing done.

Watchdog auto-compacts, restarts, and replays the last message. Truly unattended.

No visibility

Ten terminal tabs. No idea which agents are working, stuck, or idle.

Live dashboard with status, token spend, peek mode, and mobile access from anywhere.

Duplicated work

Agents step on each other. Same task gets implemented twice.

Kanban board with atomic SQLite task claiming. One agent, one task, no collisions.

Can't coordinate

Agents are isolated. No way for one to delegate to or check on another.

REST API injected at startup via $AMUX_URL. Agents send tasks and peek each other's output with a single curl call.

Dashboard

Your fleet at a glance

Session cards, live terminal peek, kanban board, and a mobile PWA — all in one place.

Session Dashboard

All running agents at a glance -- live status, token spend, model, and current task.

New

White-label it. Make it yours.

Replace every trace of amux with your own brand — name, icon, logo, and accent color. One click, server-side, synced to every connected client.

Custom name & tagline

Replace “amux” everywhere — header, page title, PWA manifest, About panel. Your team sees your brand, not ours.

Icon & logo upload

Upload a square icon (favicon, PWA, home screen) and a wide logo (About panel). Drag-and-drop or file picker. PNG, JPEG, SVG, WebP.

Accent color

Pick any color — buttons, tags, status indicators, and the entire UI theme update instantly across all tabs and devices.

Server-side, synced everywhere

Branding is stored on the server and served to every client — desktop browser, mobile PWA, iOS app. Set it once, done.

API: GET/POST/DELETE /api/branding

How it works

Built for running agents at scale

Every piece exists because we hit the same wall running dozens of agents overnight.

Self-healing watchdog — context low → auto-compact; thinking-block corruption → restart + replay; stuck waiting → auto-respond. All silent, all night.

Agent-to-agent orchestration — $AMUX_SESSION and $AMUX_URL injected at startup. A shared global memory file gives every agent the full REST API. Agents delegate work to each other in plain English.

Atomic task claiming — POST /api/board/:id/claim is a SQLite CAS. Dozens of agents race for the same queue with no Redis, no duplicated work.

Mobile PWA — installable on iOS and Android. Background Sync replays queued commands on reconnect. Manage your fleet from your phone.

Single Python file, zero build step — server + full dashboard as inline HTML/CSS/JS. Edit it; it restarts via os.execv on save. Only deps: python3 and tmux.

Status without hooks or patches — parses ANSI-stripped tmux output to detect working / waiting / idle. No modifications to Claude Code.

White-label branding — replace the name, icon, logo, and accent color. Stored server-side, synced to every connected client. See details ↑

How it's different

Not another Claude wrapper

Claude Code gives you one agent per terminal. amux gives you a fleet.

	amux	Claude Code	Other wrappers
Parallel agents	Unlimited, isolated	One per terminal	Limited or manual
Self-healing	Watchdog auto-compacts + replays	Crashes on context overflow	Basic restart only
Mobile management	Native iOS app + PWA	Terminal only	Not available
Agent coordination	REST API + atomic task claiming	None	Shared filesystem at best
White-label	Full branding: name, icon, logo, colors	Not available	Not available
Setup	git clone + ./install.sh, 0 deps	pip install	Docker, config, API keys

Run dozens of Claude Code agents from your phone