native remote session handoff for Codex
Open 💬 3 comments Opened Apr 26, 2026 by juniormartinxo
💡 Likely answer: A maintainer (github-actions[bot], contributor)
responded on this thread — see the highlighted reply below.
What variant of Codex are you using?
CLI
What feature would you like to see?
Please add native remote session handoff for Codex, so I can continue the same active coding chat/session from another computer without SSH, VS Code Remote, tmux, or screen. Similar to Claude Code remote sessions: persistent agent state, workspace continuity, and secure account-based access.
Additional information
_No response_
3 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
I would strongly support this as a first-class workflow.
The most useful version for me would be a direct bridge from a local Codex CLI session into Codex Cloud, especially after the CLI has helped produce a concrete plan. Today the local flow is useful for interactive discovery: inspect the repo, reason through the approach, refine the plan, and decide what should happen. But once the plan is clear, the actual execution often belongs in the cloud, because it may involve longer-running edits, tests, or parallel work that should not tie up the local terminal.
Concretely, I would love to see something like a slash command:
or an option after plan generation alongside the existing local continuation flows:
The key thing is preserving the intent and enough execution context, not necessarily migrating every token of the local session. A minimal version could package the approved plan, repo/branch/commit metadata, relevant instructions, and the working tree state policy, then launch a Codex Cloud task against the same repository.
This would make the CLI and Cloud feel like one agent workflow: local for high-bandwidth planning and steering, cloud for autonomous execution.
Strong +1. For this to work well, I think the hard part is more than moving the transcript between machines.
A useful handoff needs to separate durable session history from current operating state: active plan, repo/worktree identity, pending human decision, last command/result that was actually observed, and what needs to be rechecked after resume. Otherwise a resumed cloud session can inherit stale confidence from the local session instead of a verifiable starting point.