One app, every machine
Agent work does not stay on one computer. You prototype on the laptop, run heavy computation on the dev server, and test against staging infrastructure. MuxAgent pairs all of those machines to a single mobile app so you can delegate, monitor, and approve work regardless of which machine is doing it.
How multi-machine works
Relay topology: phone ↔ relay ↔ N machines
The relay server sits at the center of a star topology. Your phone connects to the relay. Each machine runs its own daemon, which also connects to the relay. The relay routes encrypted messages between any phone–machine pair. This means machines do not need to be on the same network, do not need open inbound ports, and do not need to know about each other. A laptop behind home NAT, a cloud VM in us-east-1, and a GPU server in your office closet all connect outbound to the relay and become equally reachable from your phone.
One daemon per machine, independently managed
Each machine runs its own muxagent daemon process. The daemon manages its own agent sessions, persists session metadata locally, and maintains its own WebSocket connection to the relay. Machines are paired individually — you scan a QR code per machine, and each machine generates its own X25519 keypair. In the mobile app, each machine appears as a separate entry with its own hostname, connection status, and session list. You can start a task on your laptop and a different task on your dev server simultaneously, approving each independently.
Session persistence across reconnections
A WebSocket disconnect does not lose your agent sessions. Session data is persisted on both sides: the CLI stores session state in local storage, and the mobile app stores session metadata (ID, machine ID, working directory, title, status, cost information) in SQLite. When the connection restores, the daemon reconnects to the relay, re-establishes the encrypted channel via a fresh X25519 handshake, and resyncs events. A running agent task continues uninterrupted on the machine — the disconnect only affects the visibility channel to your phone, not the work itself.
Machine pairing is physical and cryptographic
Adding a new machine takes about thirty seconds. Start the daemon, which prints a QR code to the terminal. Scan it with the MuxAgent mobile app. The QR encodes the machine's signing key fingerprint, which the app stores as the root of trust for that machine. The relay creates the machine record only when you approve on the phone — there is no window where an unverified machine can register. After pairing, the machine appears in your machine list and is ready to receive agent tasks. Revoking access is a single action in the app.
Where multi-machine pays off
Prototype on laptop, ship from dev server
Start an agent task on your laptop to draft the implementation plan. Once the plan is approved, move the execution to the dev server where the full test suite and database are available. Both machines are in your mobile app — you review the plan from one, approve execution on the other, and monitor both from the same screen. The agent work follows the right machine for each stage instead of being pinned to wherever you happened to start.
GPU compute on a remote server
When an agent task involves ML training, large builds, or heavy data processing, you want it running on the machine with the right hardware. Pair the GPU server, delegate the task, and monitor from your phone while you continue other work on the laptop. The relay handles the routing regardless of network topology — the GPU server can be in a different datacenter, behind a VPN, or on a different continent.
Team machines with shared visibility
Pair multiple team machines to your app during an incident or a coordinated deployment. You can see which machines have active agent sessions, what stage each session is in, and approve checkpoints across all of them without SSH-ing into each one individually. This is the control plane model that replaces the manual SSH-and-tmux approach once you're working across more than one or two machines.
Related reading
- From laptop to dev server: running agent work across multiple machines
The practical challenge of choosing the right machine for each stage and keeping workflow control consistent when work moves between environments.
- How to pair MuxAgent on a second machine without friction
Step-by-step guide to adding a new machine to your setup — from daemon start to QR scan to first task.
- MuxAgent vs. manual SSH and tmux
When the manual approach works and where it breaks down — a fair comparison of managing agents through SSH versus a dedicated control plane.
Add a machine in 30 seconds
Install the CLI, run muxagent daemon start, and scan the QR code with the mobile app. Your new machine is paired and ready for agent tasks. See the getting started guide for the full walkthrough.