Agents are compute-hungry in a way typing never was. A single busy session means a type-checker, a test suite, a dev server, and a build — often several times per turn. Multiply by five parallel sessions and your laptop is a hair dryer with a keyboard. Meanwhile there's probably a machine in your life with cores to spare: the desktop at the office, the homelab box, a cloud VM — quite possibly the only machine the code is even allowed to live on.
So hang4r sessions can run there. Pick SSH as the environment when starting an agent, and everything executes on the remote host — the agent CLI, the repo, the tools, the hooks, the dev server. Your Mac keeps the same tiled UI: chat, editor, terminal, diff review. The session is remote; the experience isn't.
Everything moves, not just the agent
Remote done halfway is worse than local: an agent running here against files mounted from there breaks every tool it shells out to. hang4r takes the Cursor-style position — the work happens where the code is:
- The agent runs on the remote host, streaming into your session like any other — permissions, model switching, interrupts included.
- Terminals are real PTYs on the remote machine — resize, colors, ctrl-C, the lot.
- Git and the file explorer read the remote repo, so diffs, checkpoints, and the editor show what's actually there.
- Dev servers tunnel back. The agent starts something on the remote's port 3000, and hang4r opens a per-session forward so it's browsable from your machine — the remote's localhost, on your screen.
Your ssh, not our ssh
hang4r doesn't reimplement SSH; it shells out to yours. Which means your
~/.ssh/config just works — aliases, keys, ProxyJump chains,
1Password or agent-forwarding setups — the exact configuration you already trust in a
terminal. Add a host in Settings → Remote (any target your ssh resolves),
and hang4r keeps one authenticated master connection per host so every subsequent
operation rides the same socket: no re-auth, no per-command handshake lag. Commands run
through a login shell on the far side, so nvm, Homebrew, and friends resolve exactly as
they do when you SSH in yourself.
The subscription rule doesn't bend
hang4r's founding constraint — your credentials never leave your machines — holds over
SSH in the strictest way: the remote host's CLI must be installed and logged
in. hang4r never copies a token, never proxies an API call; "Test connection" in
Settings checks reachability and that claude answers on the far end
before you ever start a session. It's the same deal as local, relocated: your machine,
your subscription, your code — just a different "your machine."
What it's for
- The big-machine dispatch. The monorepo with the six-minute build lives on the workstation. Dispatch sessions to it from the laptop; the fans that spin are someone else's.
- Code that can't come to you. Data-adjacent repos, GPU boxes, machines inside a VPN — the agent goes to the code instead of the code coming out.
- The full chain. Sessions on the beefy box, hang4r on the Mac, the phone app in your pocket — three tiers, one loop, and the compute always lives where it's cheapest.
The fine print
The remote needs the agent CLI installed and logged in — that's the point, not a
gap. And the hang4r browser assertion CLI stays local-only, since a
remote host can't reach the socket that drives your embedded browser; remote web work
verifies through the tunnel instead.