Concierge Want it run for you? Done-for-you automation for your business, from your phone · dental, skilled nursing & more Learn more →
self-hosted · open source · MIT

Run an AI engineering team
from one dashboard.

amux is the open-source control plane for AI coding agents. Indie hackers, solo builders, and engineering teams use it to launch, coordinate, and recover dozens of parallel Claude Code sessions from a single interface — with self-healing built in, so you wake up to finished work instead of crashed sessions.

~/dev — bash cmd · c
# install (or: pipx install amux · or clone the repo)
$ brew install mixpeek/amux/amux
$ amux register myproject --dir ~/Dev/myproject --yolo
$ amux start myproject
$ amux serve   # → https://localhost:8822
// Why amux

Multi-agent infra, without the duct tape.

If you're already running parallel coding agents from a stack of terminal windows, you've felt every one of these. amux fixes them in a single Python file.

Problem
amux's solution
Agents crash at 3am from context compaction
Self-healing watchdog — auto-compacts, restarts, replays the last message.
Can't monitor 10+ sessions from one place
Web dashboard — live status, token spend, peek into any session.
Agents duplicate work on the same task
Kanban board with atomic task claiming via SQLite CAS.
No way to manage agents from your phone
Remote control iOS app + PWA — monitor and manage your team from anywhere. Download on App Store →
Agents can't coordinate with each other
REST API + channels — send messages, peek output, claim tasks across sessions.
No persistent knowledge between sessions
Shared notes & memory — markdown documents agents read and write across sessions.
Dev server stuck on localhost — can't share with clients or test webhooks
Cloud tunnel — stable public HTTPS URL, same address across restarts, no firewall config.
// Dashboard

One pane of glass for the whole fleet.

Session cards, live terminal peek, kanban, notes, CRM, scheduler, files. Everything lives in a single Python file with inline HTML — clone it, edit it, deploy it.

https://localhost:8822/sessions
workspacesessions all ⌈ K
Active
8 / 12
Tokens today
4.7M +612k cache
Spend (24h)
$18.40
Tasks shipped
23 this week
workingapi-refactor
Splitting auth middleware into per-route guards. Touching 14 files.
tmux:0421k tok2m ago
needs inputmigrations
Drop legacy `users.role_id` column? Will require backfill from `permissions`.
tmux:1198k toknow
workingtest-gen-frontend
Generating Playwright e2e suite for checkout flow. 12 / 34 specs done.
tmux:21.1M tok4m ago
idledocs-sweep
Updated 8 README sections. Waiting for review on PR #482.
tmux:376k tok22m ago
workingscheduler-cron
Implementing `--every` flag with timezone-aware parsing.
tmux:4312k tokjust now
idlebug-PROJ-219
Race condition in board claim — repro pending. /compact ran 3m ago.
tmux:5540k tok3m ago
// Mobile

Your AI engineering team. In your pocket.

The remote control for your AI engineering team. Monitor live sessions, approve actions, inspect terminal output, and recover stalled agents — from anywhere. PWA for iOS/Android plus a native iOS app.

Download on the App Store
9:41
a
Sessions 8 ACTIVE · 12 TOTAL
tailnet
all working input idle
working api-refactor
Splitting auth middleware into per-route guards.
421k2m
needs input migrations
Drop legacy users.role_id column? Will need backfill.
198know
working test-gen-frontend
Generating Playwright suite for checkout. 12 / 34.
1.1M4m
idle docs-sweep
Updated 8 README sections. PR #482 awaiting review.
76k22m
Sessions
Board
Channels
Settings
9:41
«
api-refactor WORKING · 421K TOK
——— tmux peek · last 50 lines ———
reading src/auth/middleware.ts
reading src/routes/users.ts
Found 4 routes using authMiddleware
Now extracting per-route guards…
editing src/auth/guards/admin.ts
+ export const requireAdmin = ...
editing src/routes/admin.ts
+ import { requireAdmin } from '../auth/guards'
14 files staged
running tests…
PASS src/auth/guards/admin.test.ts
PASS src/routes/admin.test.ts
send to session…
Sessions
Board
Channels
Settings
9:41
Board 14 OPEN · 7 IN PROGRESS
todo doing review done
PROJ-12 claimed · api-refactor
Refactor auth middleware into per-route guards across 14 files
P1backend4h
PROJ-18 claimed · test-gen
Playwright e2e suite for checkout flow
P2frontend2h
PROJ-21 claimed · scheduler-cron
Add --every flag to amux scheduler with TZ parsing
P2cli1h
PROJ-25 unclaimed
Migrate notes index to FTS5
P3backend
Sessions
Board
Channels
Settings
// Agent ↔ agent

Agents that talk to each other.

Every session exposes a REST API. Agents discover peers, send tasks, peek terminals, and atomically claim board items — coordinating fleets of dozens without a human in the loop.

The full API reference is dropped into each agent's global memory at startup, so plain-English orchestration just works: “ask worker-2 to add tests for the changes I just made.”

orchestration.sh
# Send a task to another session
$ curl -sk -X POST \
    -H 'Content-Type: application/json' \
    -d '{"text":"implement /login and report back"}' \
    $AMUX_URL/api/sessions/worker-1/send

# Atomically claim a board item
$ curl -sk -X POST $AMUX_URL/api/board/PROJ-5/claim
{"ok":true,"claimed_by":"worker-2"}

# Watch another session's output
$ curl -sk "$AMUX_URL/api/sessions/worker-1/peek?lines=50"
// Everything inside

Batteries included.

amux ships with the workspace tools an agent fleet needs — kanban, CRM, email, browser automation, scheduler, and tunnel. Every feature lives in the same single Python file.

01

Self-healing watchdog

Monitors context %, restarts on corruption, replays the last message, unblocks stuck prompts. Designed for unattended overnight runs.

02

Parallel sessions

Run dozens of agents at once. Each gets a UUID that survives stop/start. Conversation fork clones history onto a new branch in one command.

03

Agent orchestration

REST API + shared global memory. Agents discover peers, delegate work, claim tasks, and coordinate via channels with @mentions.

04

Kanban board

SQLite-backed with auto-generated keys, atomic CAS claiming, custom columns, and iCal sync. No two agents pick up the same ticket.

05

Notes & CRM

Markdown notes with rich editor and find-in-page. Contacts, companies, interaction logs, and follow-up tracking — all readable to agents.

06

Scheduler

Named cron-style recurring jobs with a built-in management UI. Run agents on a schedule — daily reviews, nightly test sweeps, hourly triage.

07

Browser automation

Shared Playwright instance with saved auth profiles, screenshots, and an AI agent mode for tasks that need a real browser.

08

Skills & slash commands

Project-level commands like /commit or /review-pr that any agent can invoke. Define once, share across the fleet.

09

Single-file architecture

One Python file with inline HTML/CSS/JS. No build step. No dependencies. Edit and save — it restarts itself. Audit the whole thing in an afternoon.

10

Cloud tunnel cloud

Expose any localhost port at a stable public HTTPS URL — no inbound port, no firewall config. URL is token-derived so it never changes across restarts.

// growth

291 stars and growing

amux shipped in early 2026 and grew to 291 GitHub stars in months. Every star comes from a developer who found it solved a real problem.

amux GitHub star history chart — growth from 0 to 291 stars since early 2026
⭐ Star on GitHub
// ship at night

Stop babysitting one agent.
Start running the fleet.

Self-host the open-source build, or skip the setup and try amux Cloud free for 14 days.