Claude Code resume: how to continue any session

There are four ways back into a conversation. All of them read the same saved transcript, so the conversation continues with its history and on the model it was using.

claude -c                       # last conversation in this folder
claude -r                       # pick from a list
claude -r auth-refactor         # by name
claude -r 3f2a9c1e-...          # by session id

A conversation you leave with /clear is not lost either. Claude Code starts a fresh session and keeps the old one on disk, so it shows up in /resume.

The picker is the history view. Run claude --resume with no argument, or /resume, and each row shows the session name if you set one, otherwise a generated title or the first prompt, plus time since last activity, git branch and file size.

• Up and Down (or k and j) move between sessions. Enter resumes the highlighted one, and 1 to 9 resume by position.
• Space previews the conversation without opening it.
• / or any other printable key starts a search. Pasting a pull request URL finds the session that created it.
• Ctrl+A shows sessions from every project on this machine. Ctrl+W shows all worktrees of the repository, and Ctrl+B filters to the current git branch.
• Ctrl+R renames the highlighted session. Esc leaves search or the picker.

To keep a conversation as readable text rather than a transcript, open it and run /export, which copies it to the clipboard or writes it to a file you name. If you want the raw files and their layout, that is covered in where Claude Code stores sessions.

Generated titles work, but a name you chose is easier to find and type. Set one at launch with -n, rename later with /rename (alias /name), or press Ctrl+R in the picker.

claude -n auth-refactor
# later, from the same repository
claude --resume auth-refactor

Name resolution covers the current repository and its worktrees. The short default label you see on a running session, such as my-app-3f, is not a resume handle.

Claude Code stores sessions per project directory, which is why claude --continue in the wrong folder finds nothing. There are two ways round it:

• By id, from anywhere. claude --resume <session-id> looks in the current project and its worktrees first, then every other project on the machine. If no transcript matches, it prints No conversation found with session ID: <session-id>.
• From the picker. Press Ctrl+A and pick the session. A session from another worktree of the same repository resumes in place. One from an unrelated project puts a cd plus resume command on your clipboard instead.

Sessions only exist on the machine that ran them. Moving one to a second computer is its own task, covered in continue a Claude Code session on another computer.

Resuming the same session in two terminals interleaves both into one transcript. When you want to try a different approach and keep the original intact, fork it: --fork-session gives the resumed copy a new session id.

claude --continue --fork-session
claude --resume auth-refactor --fork-session

Inside a session, /branch does the same at the current point and switches you into the copy. The original stays in the picker under its old id.

Non-interactive runs resume the same way. Pass a session id to send a follow-up and capture the answer:

claude -p --resume <session-id> --output-format json "summarize what we changed" | jq -r '.result'
claude -p --continue "run the tests again"

Sessions created with claude -p or the Agent SDK are left out of the picker and out of a plain claude --continue. Resume them by id, or use claude -p --continue, which includes them. That gap is the usual reason a session seems to be missing; Claude Code session missing from the list walks through the other causes. To run without saving anything at all, add --no-session-persistence to a -p run.

Claude Code deletes transcripts older than 30 days. If you rely on resuming old work, raise the limit with cleanupPeriodDays in ~/.claude/settings.json. The minimum is 1; there is no value that means forever, so use a large number.

{
  "cleanupPeriodDays": 365
}

To delete history now rather than wait, use claude purge. It removes a project's transcripts, tasks, file history and config entry.

claude purge --dry-run     # list what would be deleted for this folder
claude purge               # delete it, after a confirmation
claude purge --all         # every project on this machine

-i asks before each item and -y skips the confirmation. Purge is not undoable, so run the dry run first.

The resume picker shows one machine and, by default, one folder. Termdeck reads the same transcript files through a small agent on each of your machines, so the Claude Code sessions in every project folder you add, on every machine, sit in one sidebar grouped by folder. Sessions you started in a terminal are listed too, unless you turn that off in Settings. Open one and type to continue it; Termdeck resumes the same session id, so claude --resume in the terminal later picks up where the browser left off.

If a terminal is still attached to the session, the chat opens view only rather than letting two writers fork the transcript. The details are in the sessions docs. Transcripts stay on your machine.