What survives when I detach from a session?

If the host server stays alive, the original shell, agent, and other processes can keep running. Reattach reconnects the client to that live session.

Does session persistence mean the process survives a restart?

Usually no. A tool can restore layout and saved screen state, and a supported agent integration can resume a native conversation. An arbitrary process must be restarted or recovered.

Will an agent keep working while my laptop sleeps?

No. Sleeping pauses execution. Keep the host awake or move the runtime to an always-on machine. Pane Chat resilience can help continue a failed turn after wake, but it can't run code in the dark.

What does Herdr restore after a server restart?

Herdr restores the saved workspace shape and can resume supported agent sessions that reported a native reference. Its docs distinguish that from the original processes, which don't survive a full server restart.

Do persistent cloud volumes keep agents running?

No. A persistent volume keeps files, repos, and configuration. Stopping the VM stops running processes. Start it again and recover the work.

plain-language glossary

what is session persistence?

TL;DR

Session persistence means coming back to work without rebuilding it by hand, but the term covers different guarantees. Detaching keeps live processes running; a restart brings back your layout and lets supported agents resume their conversations; a sleeping host runs nothing; and a stopped cloud VM keeps its disk but not its processes. After a restart, Pane restores your panes, tabs and terminal output, and Claude Code, Codex, Cursor and OpenCode resume their conversations.

Download*

Windows SmartScreen warningDirect downloads can show a SmartScreen warning while Pane is unsigned. Pane is fully open source, so you can audit the code and build from source yourself.1. Click More info2. Click Run anyway3. Continue the installerThe PowerShell install downloads the official release directly and avoids most browser download friction.npm global install

Paste that in PowerShell.Linux

Session persistence means you can come back to work without rebuilding everything by hand. That sentence hides several different guarantees. A detached live process, a saved workspace layout, an agent conversation restored after a restart, and a persistent cloud disk are different things. Know which one your tool actually provides.

four persistence cases

Four different promises hide behind one phrase. Know which one you're getting.

case	what remains	what to expect
detach and reattach	client leaves; host keeps the server and processes alive	Herdr's strongest path; original processes keep running
app or server restart	your panes, tabs and terminal output come back; claude code, codex, cursor and opencode resume their conversations; the agent processes themselves start fresh	layout returns; supported agents can resume native sessions
host sleep	nothing executes while the host sleeps; use qualified awake-host setup	processes pause until the host wakes
cloud VM stop/start	repos, worktrees, and config stay on the VM's disk	disk data returns; running processes don't

detach is the easy case

Detach keeps the real process alive, as long as the host stays up.

Detach means the client stops displaying the session while the host keeps it alive. tmux, Zellij, and Herdr use this model. You can close a terminal or lose an SSH connection, then reattach while the original shell or agent is still running.

→ the host stays awake
→ the server stays running
→ the original process keeps its memory and state
→ reattach shows the live session

restart means recovery

After a restart you get your layout and, for supported agents, the conversation, not the old process.

When an app or session server restarts, the old process may be gone. A tool can still restore the saved workspace shape, working directory, focus, and recent screen. Some agents expose their own session IDs, so an integration can resume the conversation. A generic shell or unsupported process starts fresh.

→ layout restore isn't process restore
→ screen history isn't a live shell
→ native agent resume depends on the agent and integration
→ a crashed agent can still need a human or an orchestrator to continue it

sleep is a host power question

A sleeping computer runs nothing. Keep it awake or move the work.

No session layer can make a sleeping computer run code. If you need unattended progress, keep the host awake with a qualified power setup or use an always-on machine. Pane keeps your computer awake by default while sessions are active, and documents idle-sleep and lid-closed setups separately. Its optional Pane Chat resilience can resume a turn that died during a sleep or network error, but that's recovery, not running code while the machine sleeps.

→ idle-sleep prevention is different from lid-closed setup
→ battery sleep can pause work
→ a crash or lost network can end an agent turn
→ recovery tooling can continue a task after the host returns

cloud storage isn't process immortality

A persistent disk keeps your files, not your running processes.

A cloud VM's disk can keep repositories, worktrees, configuration, and history across a VM stop and start. That protects data. It doesn't keep a shell, test, or agent process alive while the VM is stopped. Start the machine, reconnect the client, and recover the session when needed.

→ volume persistence protects files
→ an always-on VM protects live execution
→ snapshots help before destructive changes
→ remote clients still depend on a reachable host

how to choose the guarantee you need

Match the guarantee to the gap: a network drop, a restart, a night away, or a stopped VM.

Use this quick check before you leave an agent running.

1For a short network drop, use a live server and reattach. Herdr, tmux, and Zellij are built around this path.
2For a laptop restart, check whether your tool restores layout, screen history, or the agent's native conversation.
3For overnight work, keep the host awake while plugged in or move the runtime to an always-on remote machine.
4For cloud stop/start, verify volume persistence and plan to restart processes after the VM returns.
5For a failed agent turn, inspect the last output and continue only after you know what the agent was waiting on.

Sources

Corrections

If anything here is wrong or out of date, tell us and we'll fix it and update the date at the top. Open an issue on GitHub

Sources

frequently asked questions