Sessions

Grok saves every conversation to disk automatically — prompts, responses, tool calls, and file snapshots — under ~/.grok/sessions/, keyed by working directory. Sessions work the same in the TUI, headless mode, and over ACP.

In the TUI, /resume opens a picker of recent sessions for the current workspace; the welcome screen lists them too. From the command line:

grok --resume <session-id>   # a specific session
grok --resume                # the most recent for this directory
grok -c                      # shorthand: continue the most recent

In headless mode, read the session ID back from JSON output and pass it to -r to build multi-step automations:

grok -p "Start the refactor" --output-format json | jq -r '.sessionId'

-s, --session-id names a new session with a UUID you supply; it does not resume existing ones. To branch a resumed session instead of continuing it, add --fork-session.

/fork [directive] branches the current session into a peer that starts from a copy of the conversation. Pass --worktree or --no-worktree to choose whether the fork runs in an isolated copy of the repository, so parallel sessions do not overwrite each other's files — see Worktrees.

/rewind (or Esc Esc while idle) lists a rewind point per prompt. Selecting one restores all files to their state at that point and truncates the conversation to match. Rewind modifies files on disk — reverted changes are lost unless committed to git.

/compact [context] compresses the conversation history to reclaim context window, with optional instructions about what to preserve. Grok also auto-compacts as the context window fills; check usage with /context or /session-info.