Claude Code supports /resume VPS workflows over SSH and staging servers

• levelsio's VPS workflow post describes running Claude Code on a Hetzner VPS, staying connected through SSH in Termius, and relying on Mosh, tmux, or /resume so sessions survive laptop sleep and disconnects.
• The remote setup shifts the work off the laptop, which levelsio says improves battery life, and his follow-up reply adds that Claude Code should still work on a basic VPS because the CLI is mainly a frontend for requests to Claude's servers.
• Anthropic's new multi-session control plane is part of the same shift: trq212's launch note calls Agent view a native way to manage multiple Claude Code sessions, while Cat Wu's usage tip shows the claude agents command and how sessions register into it.
• The rough edges are still mundane ones. levelsio's screenshot-paste complaint asks for automatic image upload from Termius to a server temp folder, and hckmstrrahul's lost-progress post shows what happens when a long local run dies with a Mac restart.
• Visibility is another open problem. Peter Yang's hang report showed Claude Code sitting for more than three minutes with little feedback, and Boris Cherny's reply said Anthropic was working on more responsive UX and debug logs.

Anthropic's Claude Code overview frames the tool as a terminal coding agent, while levelsio's setup shows what happens when you treat that terminal as an always-on server instead of a laptop app. You can also see the new control-plane direction in trq212's Agent view post and Cat Wu's command example, which turns parallel sessions into something closer to a queue than a pile of tabs.

The practical trick is simple: keep Claude Code on the machine that stays awake. In levelsio's main post, that means a Hetzner VPS plus SSH through Termius, with Mosh, tmux, or /resume handling disconnects.

His thread turns that into a concrete remote loop:

• file a bug or feature from the phone or laptop
• let Claude Code work on the server
• auto-commit changes to GitHub
• keep staging or deploy steps on the same remote box

The caveat is in levelsio's follow-up: he edits production directly, but explicitly says the safer version is a staging VPS that stays on 24/7.

The new claude agents flow makes the SSH-first setup more interesting because it gives those remote sessions a control plane. trq212 describes Agent view as a Claude Code native way to manage multiple sessions, and Cat Wu's command example says the intended pattern is to launch it from a top-level code directory, then register CLI sessions with the left-arrow shortcut.

That is a different bottleneck than the old one-terminal-per-task habit. Boris Cherny's post summarizes the pitch as leveling up from one agent to many without cycling between terminal tabs, while Aakash Gupta's commentary argues the change is really about supervising several work streams from one pane.

The most obvious missing piece in the VPS workflow is image handoff. levelsio's request asks Termius to detect pasted images, upload them over SFTP to /tmp, show a progress bar, and drop the remote filepath straight into chat.

The other pain point is session durability when work still runs locally. hckmstrrahul's post asks how to recover after a Mac restart killed a long Claude task, which is exactly the failure mode the server-side setup is trying to avoid.

Even when the session survives, the interface can still feel opaque. Peter Yang's screenshot showed Claude Code appearing stuck for 3 minutes and 10 seconds without enough feedback to tell whether it was still doing useful work.

Anthropic acknowledged that one directly. In Boris Cherny's reply, Cherny said the team wanted the UX to feel more responsive and planned to add debug logs so users could see what was causing hangs.