This is actually the thing that drove me to the GUI. You can just setup connection via ssh without thinking about it. A lot of my projects either live in a VM or VPS and so using herdr was okay, but the latency was getting to me and swapped to Codex and Claude Code desktop which make it super easy to just add remote machines over ssh and then there is no latency etc.
But you can't add a VPS to codex remote, can you?
Yes you can. You start a server with codex cli. Do a round trip with an authorisation token and you are in.
I couldn't manage to get it working with my vps and the android chatgpt app. Still using ssh and more comfortably pi remote with port forwarding.
You can run claude/gemini daemons on remote boxes and then use claude desktop app or antigravity or even just the web browser to drive the workflow from your preferred endpoint (laptop or even mobile phone). For me personally this is the best of the both worlds, especially considering that daemons run on env that is close to production, and my client is just a view into it, and GUI part of it allows for more efficient and more rich experience of going through the requirements, spec writeups, code generation and finally experiments. Writing, compiling, debugging, testing or profiling the code manually is no longer done by me directly, and this was the main use-case for me previously preferring and using TUI based tools such as tmux+vim.
Claude code in the Claude desktop app can run in cloud containers. I actually use the desktop app because it’s easier for sandboxing and headless.
You can also use Claude Code TUI on a remote box with /remote-control and use the GUI to control it.
This. I can't imagine trusting any of these models on my local machine.
You can do the same with a GUI if you write a client-server architecture