Most of us have more than one machine that could run an AI coding agent. A desktop with the big GPU, a homelab box that is always on, a work laptop that holds one repo nobody else has. The agent ends up running wherever you happened to open the terminal, and the session is stuck there: close the lid and it stops, walk to another room and you cannot see it.
Disclosure: I am on the Clusto team, and this post shows how Clusto handles several machines. The idea behind it (run the agent next to the code, keep the session where it runs, reach it from any device) works with plain SSH and tmux too, just with more glue.
One daemon per machine, one mesh
Clusto puts a small daemon, clustod, on every machine you want to run agents on. On Linux it is one line plus a short wizard:
curl -fsSL https://dl.clusto.app/install.sh | sh
Then you join the machines into a cluster from the app: Cluster > Adopt Node, paste a pairing code you generated on the new node, done. No config files to copy, and the node you adopt from checks the new node's identity before it sends anything.
The cluster is leaderless. Every node applies its own changes immediately and gossips them to its peers, so there is no election and nothing breaks when a node is off. Up to about five nodes form a full mesh. When a machine comes back, the nodes exchange snapshots and catch up.
What you get in the app is one view of every machine: a card per node with live CPU, memory, swap and network, refreshed about every 10 seconds, and the sessions running on each.
Start the agent where the code is
When you open a new chat or terminal session you pick the node it runs on. The New chat dialog has a Node picker, and you can set a default node for every new chat. Claude Code, Codex, OpenCode and OpenRouter models are all available from the same dialog, as long as the CLI is installed on that node.
The process runs on that machine. Clusto does not migrate processes between nodes, and that is deliberate: the agent needs the repo, the toolchain and the credentials of the machine it runs on. What moves is you. Every node serves the same session list, so you can open the desktop app, the Android app, the web app or the CLI and attach to the session on whichever node owns it. The terminal keeps its scrollback, a chat keeps its transcript and any pending permission prompt.
If a node goes offline, its sessions show as unreachable instead of disappearing, and come back when the node does.
Reaching machines you cannot reach
The awkward part of a multi-machine setup is the network. The homelab is behind your home router, the work laptop is behind a corporate NAT, and the phone is on mobile data. The usual answers are port forwarding, a VPN, or Tailscale on every device.
Clusto Tunnel is the paid answer to that (it needs a Pro license). Turn it on once for the whole cluster in Settings > Cluster. Desktop and mobile apps then connect to nodes peer to peer over QUIC with hole punching, end-to-end encrypted, and a relay only helps set the connection up. The web app cannot hole-punch from a browser, so it falls back to a relay.
The tunnel also carries the cluster mesh itself, so a laptop that leaves your network stays a full member of the cluster. Only the machine being dialed needs the tunnel enabled.
One thing it does not change: the tunnel adds reachability, not access. Every connection still authenticates with the node's own token, exactly like a direct connection.
If you already run Tailscale or WireGuard, you do not need the tunnel at all. Clustering and remote sessions work on the free Personal plan over any network that can reach the nodes.
What this looks like day to day
- A long refactor runs on the always-on box overnight. In the morning you check it from your phone and answer the permission prompt it stopped on.
- A quick fix in the work repo starts on the laptop, because that is where the repo and its credentials live.
- The cluster view shows which machine is busy before you start another agent on it.
A 26-second walkthrough of the whole flow, recorded from the real app: https://mastodon.social/@clusto/117432805503724417
The free Personal plan covers one user and three devices, including clustering. More at clusto.app, and the details are in the clustering docs and the Clusto Tunnel docs.
Top comments (0)