Support multiple remote Codex machines/runtimes in one client

[johnnykor82]

Support multiple remote Codex machines/runtimes in one client

Summary

Codex already has remote-control concepts: the mobile ChatGPT app can supervise Codex running on a computer, and the CLI exposes app-server/remote primitives such as codex app-server, codex remote-control, and codex --remote.

However, the current product experience appears to be tied to a single connected computer/runtime. I would like Codex clients, especially the mobile and Desktop clients, to support multiple connected local machines/runtimes at the same time.

Problem

I work across multiple physical machines/servers. For example:

• MacBook A has one set of local projects, credentials, services, and files.
• MacBook B has a different set of local projects, credentials, services, and files.
• Both machines can run Codex locally and are reachable through a private network such as Tailscale.

Today, the ChatGPT mobile app can remotely control Codex on one computer, but I cannot use one app as a unified control plane for multiple connected computers. I still need to switch computers, open remote desktop, or manage separate local sessions manually.

This is a limitation because Codex agents are most useful when they run close to the real local environment: local repositories, databases, credentials, running services, hardware, and project-specific tools.

Requested feature

Add multi-machine / multi-runtime support to Codex remote control clients.

The client should expose a machine/runtime switcher, for example:

• MacBook Pro
• Work Mac
• Home server
• Cloud dev box

For each machine, show:

• online/offline status
• active Codex sessions/threads
• workspace/project path
• current task status
• pending approvals
• recent command/tool activity
• model/config profile
• available skills and MCP servers
• whether the runtime is local, LAN, Tailscale/private network, or cloud

Expected behavior

A user should be able to:

1. Register multiple computers/runtimes with the same Codex account.
2. See all connected machines from the same mobile/Desktop client.
3. Start a Codex task on a specific machine.
4. Resume an existing task on a specific machine.
5. Approve or reject commands for the selected machine.
6. Clearly see which machine will execute filesystem and shell operations.
7. Keep files, credentials, workspaces, MCP config, and local services isolated per machine.
8. Revoke a machine from the client.

Why this matters

Many serious Codex users work across several machines. Each machine has different repos, local services, secrets, hardware, and network access. Remote-control support is already useful, but binding the client to one computer makes it incomplete.

Multi-machine support would make Codex a real orchestration layer for local agents, not just a remote viewer for one computer.

This would also reduce the need for unsafe third-party “remote Codex UI” tools. There have already been reports of malicious packages masquerading as Codex remote/web UI tools and stealing authentication tokens. A first-party multi-machine remote control UX would be much safer.

Security requirements

This feature needs explicit, conservative UX:

• prominently show the active machine before every approval
• never silently share credentials between machines
• per-machine trust settings
• per-machine MCP/server visibility
• per-machine audit log: which machine executed which command
• support Tailscale/private-network-only endpoints
• support secure token exchange and revocation
• avoid exposing local app-server endpoints publicly by default

Related existing capabilities

Codex CLI already exposes related primitives:

• codex app-server
• codex remote-control
• codex --remote <ADDR>

The requested feature is to expose this multi-runtime model in first-party mobile/Desktop clients with a safe UX.