Remote Control enrollment never completes on Windows

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• #31387
• #31947
• #31799

Powered by Codex Action

Happening in MacOS too

Update: it was happening in app after i had set two-factor auth in the web; had to log out from the app and log back in to make the issue disappear

Happening in MacOS too

I bought Macbook neo and its working fine on the same network and the same account i think its a windows serves issue

I can reproduce the same Windows host-enrollment failure on a newer build.

• Codex Desktop: 26.721.4979.0
• Embedded local app-server reported by the desktop log: 0.146.0-alpha.3.1
• Trigger: Settings > Connections > Control this PC > Allow connections > Add
• Actual result: the pairing dialog displays Pairing code unavailable: Remote control failed to connect, then the app displays Couldn't update remote control availability and switches Allow connections back off.
• The failure occurs before the iPhone can scan or enter a pairing code.
• SSH connections from the same PC work normally.
• WinHTTP proxy: direct access, no proxy configured.
• Reproduced July 27, 2026 at approximately 05:54 to 05:55 Pacific.

Local evidence: remoteControl/enable returns without an app-server error, but immediately afterward repeated remoteControl/client/list calls and refreshes produce zero remote-control clients. The PC maintains an outbound connection to OpenAI, so this looks like a remote-control enrollment/relay failure rather than a local network or phone issue.

In-app feedback was uploaded successfully:
no-active-thread-019fa3c6-f7eb-75a0-b28b-41fdc314991d

I have a 36-second screen recording showing the full failure and can provide it if needed.

Thanks. I will run the requested isolated-network test before resetting the Windows app again.

Already completed before this request:

• Confirmed the Microsoft Store Codex app is current: 26.721.4979.0.
• Repaired a malformed local Codex configuration and verified it parses correctly.
• Backed up and cleared the stale local remote-control enrollment once, then restarted Codex. The problem persisted.
• Updated the SSH-discovered VPS Codex runtime from 0.132.0 to 0.145.0. This did not restore phone pairing, and current failures occur before the phone receives a pairing code.

Current host checks:

• WinHTTP proxy: direct access, no proxy configured.
• Windows user proxy: disabled. I have no TLS-inspection proxy installed or configured.
• Codex maintains outbound TCP 443 connections to OpenAI while the failure occurs.
• Windows Firewall does not have a restrictive outbound policy configured for the active profiles.
• Tailscale is currently running as a network overlay, but it is not configured as an exit-node route. I will temporarily stop it for the hotspot test.

The remaining test is to connect the PC to an iPhone hotspot, with Tailscale stopped, then retry Settings > Connections > Control this PC > Allow connections > Add. I will report that result here before considering a Windows app Reset and reboot, since Reset clears local application data.

Result of the requested clean-network test: failed identically.

I connected the Windows PC to an Android cellular hotspot (SSID Verizon-SM-G930V-7E78, 90% signal), then ran tailscale down successfully before reproducing the failure. No proxy, VPN overlay, home router, or TLS-inspection path was active for this attempt.

At 09:55 to 09:57 Pacific on July 27, Settings > Connections > Control this PC > Allow connections > Add still showed the same pairing-code failure and turned Allow connections back off.

Fresh logs again show remoteControl/enable completing without an app-server error, immediately followed by refresh_remote_control_completed ... nextConnectionCount=0 and repeated remoteControl/client/list calls with no remote-control client. This rules out the requested network, VPN, and proxy conditions.

I restored Tailscale afterward. The remaining prescribed local step is a Windows App Reset plus reboot. I will do that next only after confirming it is acceptable because it clears the desktop app cache and sign-in state.

MacOS reproduction: Remote Control pairing unavailable after host enrollment

Environment:

• macOS 15.5 (24F74), Mac Mini
• ChatGPT Desktop 26.721.41059 (build 5848)
• Embedded app-server 0.146.0-alpha.3.1
• Standalone Codex 0.145.0 used only as a diagnostic control
• iPhone ChatGPT app signed into the same account/workspace

Reproduction:

1. ChatGPT Desktop > Settings > Connections.
2. Enable Allow connections.
3. Click Add and keep Phone selected.
4. Pairing dialog spins, then reports: Pairing code unavailable: remote control pairing is unavailable until enrollment completes.

Observed backend/app-server behavior:

• Host receives a server ID and environment ID.
• Desktop remote websocket reaches Connected and reports has_enrollment=true.
• remoteControl/pairing/start still returns enrollment incomplete.
• In a standalone-only control run, the pairing request coincides with WebSocket protocol error: Connection reset without closing handshake.

Controls already tested:

• Signed out and back into ChatGPT Desktop.
• Fully quit and reopened the app.
• Confirmed same account/workspace on Mac and iPhone.
• Closed the Windows ChatGPT client.
• Verified direct reachability to ChatGPT/OpenAI endpoints.
• Installed the official standalone Codex component.
• Tested with Desktop as the only app-server.
• Tested with standalone daemon as the only app-server.
• Removed a duplicate local app-server condition; Desktop then connected successfully, but pairing remained unavailable.
• Repeated attempts remain identical well beyond 10 minutes.

No public tunnel, proxy, or VPN is involved in this setup. Chat and other Codex functionality work normally. This appears to be a backend enrollment state where the server/environment records exist and the websocket can connect, but the pairing endpoint does not consider enrollment complete.

Fresh macOS retry after restoring a strict single-host state:

• ChatGPT Desktop remains the only app-server/Remote Control host; the standalone diagnostic daemon was stopped before this attempt.
• ChatGPT's built-in updater reports 26.721.41059 is the newest version available.
• Allow connections is enabled and Phone is selected.
• Retried at 2026-07-29T01:03:45Z.
• remoteControl/pairing/start returned error -32600: remote control pairing is unavailable until enrollment completes.
• Request ID: fe5f4473-7dad-4299-b3db-fface37c7760.
• No QR code or pairing code was issued.

This reproduces after removing the duplicate host condition, so the remaining failure is still the server-side enrollment gate rather than host contention.

I'm having exactly the same issue on Windows. With macOS it works smoothly without any issue.

I am seeing the same issue on Windows 11.

Remote Control shows:
“Pairing code unavailable: remote control pairing is unavailable until enrollment completes.”

I reset the local Remote Control state, including:

• electron-local-remote-control-installation-id
• electron-local-remote-control-environment-id
• electron-remote-control-config-migration-completed
• codex-mobile-has-connected-device

After restarting Codex, the state is recreated, but Remote Control still never enrolls.

Current desktop logs repeatedly show:
savedRemoteConnectionCount=0
totalConnectionCount=0
nextConnectionCount=0
previousConnectionCount=0
creationFailureCount=0

The app continuously logs:
refresh_remote_control_started
refresh_remote_control_completed

but never creates a Remote Control connection.

Windows 11, Microsoft Store Codex app.