How it works

A safe route from web AI intent to local Codex.

ChatGPT decides when execution is appropriate and writes a concise task. Runmote delivers it to the active local Codex session, tracks the lifecycle, and returns the agent-native report.

1

Install a Runmote package

Choose CLI stable, CLI beta/RC preview, Desktop beta, or Headless beta for the machine role. Install is passive: setup, pairing, and local runtime start only when you initiate them.

2

Pair your device

Runmote account pairing links the local client to runmote.com. Codex/OpenAI or ChatGPT authentication remains handled by those tools outside Runmote.

3

Report safe local state

The CLI can report bounded presence and workspace-shape metadata such as workspace kind and Git-root count, not secrets or raw file contents.

4

Run one agent task

Runmote checks ownership, lifecycle state, active reservations, quota, and idempotency before transporting the opaque prompt once. Codex chooses the implementation details.

5

Return and review

Runmote returns bounded state and the opaque terminal report. ChatGPT assesses that report and gives the user a concise completion, owner action, or next step.

Generic lifecycle

Start, status, continuation, cancellation, duplicate suppression, and bounded opaque reports use one canonical agent-task interface.

What stays local

Codex workspaces, Codex/OpenAI login, and local commands stay on the user's machine behind the device credential boundary.

What is deferred

Artifact viewing, preview links, browser terminals, sudo-capable terminals, public sharing, and future stable Desktop/Headless packages.