A macOS terminal for remote VM users

While training models at Pinch, I was constantly SSH’d into different GPU machines, each running tmux inside separate VS Code windows.

It worked, but I felt like the bottleneck for my agents productivity. Too heavy to context-switch and tmux scrollback issues were painful.

So I built Haven — a persistent terminal+agent manager for Mac

Link is here if you’d like to try it: https://haventerminal.com

• open source daemon for persistence that it sets up when connecting to VM
• It’s fast! GPU-accelerated terminal + predictive echo to make VM connections feel local
• native scrollback regardless of TUI (claude code, codex, etc.)

I’m fully on Haven for all my work now and feeling less cognitive fatigue (+ a lot more token usage)

I take pride in design and efficiency, so I’d love to show off some screens below.

everything you see works across local and remote sessions

Grids without the overhead of remembering where things are running

One-click machine overview (CPU/GPU usage)

⌘ + K to generate command

Haven gives Claude a canvas to show things like images, videos, specific code diffs, etc.

Filesystem and file viewer/editor

All gifs/screenshots are from version 0.2.1.

Would love feedback from anyone SSH-heavy, especially the ML folks!

https://haventerminal.com