in active development

The workspace for
parallel AI.

Run 10 agents at once. See which ones need you. Navigate with your hands. Control from your phone. gmux is the ambient layer your AI development is missing.

Get early access → ⚡ GitHub

10×

agents in parallel

<2s

state detection lag

400ms

voice latency

1:myapp! permission

⚠ Write to auth.py?
[y] once [a] always [n] deny
▌

gmux · agents◉ 4 active

myapp! perm

api-server◉ working

database● waiting

frontend◉ working

2:api-server◉ working

Refactoring route handlers▌
✓ routes/auth.ts
✓ routes/users.ts
› routes/api.ts

📷· !1· ◉2· ●3· ◉4· ^!5 tmux · live

gesture navigation · voice routing · phone control · real-time SSE state · tmux + opencode · sub-agent detection · session restore · MediaPipe hands · gesture navigation · voice routing · phone control · real-time SSE state · tmux + opencode · sub-agent detection · session restore · MediaPipe hands ·

what's included

Everything your workflow
has been missing

gmux wraps tmux. No tool replacement — just the visibility and navigation layer that makes parallel AI development possible.

🔍

Live state detection

Subscribes to every opencode SSE stream. Knows the exact moment an agent switches state — not by polling, by listening.

✋

Gesture navigation

MediaPipe hand tracking. Right hand navigates windows. Left hand commands. Swipe, point, three-finger agent jump.

🎤

Local voice routing

faster-whisper STT on-device. Nav to tmux. AI queries to the focused pane. 400ms. No cloud. No keys.

📱

Phone as remote

Mobile PWA, zero install. Agent carousel. Volume-key cycling. Push-to-talk to the focused agent.

🖥

Tauri native UI

Real xterm.js terminal via PTY, live sidebar with todo counts, one-click permission approval, gesture calibration.

♻️

Session persistence

Window names saved every 30s. Agents relaunch on resurrect. Custom names survive kills and reboots.

$ gmux status

✓ monitor.py running · SSE connected on 9 panes

✓ session_restore daemon · saving names every 30s

✓ cam-broker /dev/video2 · 30fps

· voice bridge offline · start with voice-toggle.sh

state system

Every agent, every state,
at a glance

Permission beats waiting beats working. Each window shows the most urgent pane inside it.

permission

Main agent needs tool approval now

sub-agent perm

Task-tool child needs approval (one level down)

●

waiting

Agent ready for your next input

◉

working

Running tools, streaming response

◆

done

Just completed a task

✗

error

Broke — needs your attention

○

idle

Plain shell, no AI running

gesture control

Two hands.
Two roles.

Right hand navigates. Left hand commands. The camera broker keeps gesture tracking off your browser's camera.

◉ right hand — navigation

swipe right →next tmux window

swipe left ←previous tmux window

swipe up ↑scroll up

swipe down ↓scroll down

pinchselect / configurable

◉ left hand — commands

point ☝toggle voice listen

three fingersjump to next waiting agent

thumbs up 👍accept / approve

thumbs down 👎reject / deny

peace ✌always-listen mode

early access

Be first to know

Terminal stack is working. Tauri UI is being built. Leave your email — we'll reach out when it's ready.

The workspace forparallel AI.

Everything your workflowhas been missing