Skip to content

Codex mobile intermittently fails to open running task conversations #28340

Open
Open
Codex mobile intermittently fails to open running task conversations#28340
Labels
bugSomething isn't workingconnectivityIssues involving networking or endpoint connectivity problems (disconnections)iOSIssues related to the Codex iOS appsessionIssues involving session (thread) management, resuming, forking, naming, archiving

Description

@GGBondBlueWhale
GGBondBlueWhale
opened on Jun 15, 2026

What version of the Codex App are you using?

Desktop host version not captured in the screenshot. A feedback bundle was uploaded with this ID:

019eca26-18b8-7442-a042-cd22d29d2984

Mobile client: ChatGPT iOS app / Codex mobile surface. Current client environment: iPhone, iOS 26.5, ChatGPT 1.2026.153.

What subscription do you have?

ChatGPT Pro.

What platform is your computer?

macOS desktop host shown as MacBook-Pro-4.local in Codex mobile.

What issue are you seeing?

Codex mobile cannot reliably open the conversation for an already-running task.

When tapping into an active task from the iOS Codex mobile surface, the conversation sometimes fails to load and shows:

Error loading messages: Codex connection was invalidated.

At the same time, the mobile UI shows the host as reconnecting:

MacBook-Pro-4.local is reconnecting

The affected task still appears to be active, showing:

Pursuing goal — 11h 35m

So the problem appears to be that Codex mobile loses/invalidates the connection or message-loading state for an in-progress task conversation, rather than the task simply being completed or intentionally unavailable.

Steps to reproduce

  1. Start a long-running Codex task from a desktop host.
  2. Open the same running task from Codex mobile in the ChatGPT iOS app.
  3. Leave the task running for an extended period, or return to it later from mobile.
  4. Attempt to open the running task's conversation again.
  5. Observe that the task card is still present, but the message thread fails to load with Codex connection was invalidated, while the host status remains stuck on reconnecting.

Expected behavior

Codex mobile should reliably reopen an in-progress task conversation and load its messages, even after the task has been running for a long time or the mobile connection has been interrupted.

If the host is temporarily reconnecting, the mobile app should recover automatically once the host connection is restored, without making the task conversation inaccessible.

Actual behavior

The running task remains visible, but the conversation cannot be opened reliably from mobile. The UI shows Error loading messages: Codex connection was invalidated. and the host appears to be stuck reconnecting.

This makes Codex mobile unreliable for monitoring or continuing long-running desktop tasks away from the Mac, because the user may see that the task is still running but cannot read or interact with the conversation.

Additional context

This was captured from Codex mobile while monitoring a long-running task titled in Chinese. A feedback bundle was uploaded immediately after the failure; please use the feedback ID above for logs/session details. I am not posting private task contents publicly, but can provide more redacted details if useful.

Activity

  1. github-actions
    added
    bugSomething isn't working
    iOSIssues related to the Codex iOS app
    connectivityIssues involving networking or endpoint connectivity problems (disconnections)
    sessionIssues involving session (thread) management, resuming, forking, naming, archiving
    on Jun 15, 2026

  2. jwitkoski commented on Aug 10, 2026

    @jwitkoski
    jwitkoski
    on Aug 10, 2026

    Same issue happened to me. Initiated a long-running /goal prompt via codex CLI 0.147.0 while ssh'd to the remote host. I was initially able to open the thread on iOS, but now I get this same error message.

    While on the iOS "Remote" page, ass the tasks/threads are visible (including the inaccessible long-running thread) & other threads seem to work just fine.

    Exported session: 019fe793-aecc-7a13-a30f-86aeb8e5b5e7

  3. andrewrice commented on Aug 11, 2026

    @andrewrice
    andrewrice
    on Aug 11, 2026

    A recovery path I found:

    1. Quit desktop app
    2. Quit iOS app
    3. Re-open desktop app
    4. In desktop app Settings -> Connections -> Invalidate session
    5. In desktop app: Settings -> Connections -> Add
    6. Reconnect iOS

  4. matskevich commented on Sep 4, 2026

    @matskevich
    matskevich
    on Sep 4, 2026 – with ChatGPT Codex Connector

    i can reproduce a closely related failure with two different local threads from the ios codex client.

    environment:

    • desktop app: 26.825.31414, build 7287
    • bundled codex cli: 0.150.0-alpha.12.2
    • host: macOS 26.6.2, MacBook-Pro-2.local
    • client: iphone codex mobile surface
    • local time: september 4, 2026, approximately 12:02–12:07 wita

    symptoms shown on ios:

    • “error loading messages”
    • “this task couldn't reconnect”
    • “codex request timed out”
    • an indefinite loading spinner

    affected threads:

    • 01a06a0c-4c68-70d1-a237-c48b64e08deb
    • 01a06a95-52ea-7413-aa25-a79b9eeae760

    important diagnostic detail:
    both threads were already in the idle/completed state. their turns and final responses were preserved. while ios reported that the tasks could not reconnect or load messages, the desktop app-server remained responsive. subsequent thread/read and thread/turns/list calls completed successfully in approximately 1–14 ms with no routing error.

    this suggests a stale or invalid ios hydration/reconnect state rather than an unavailable host, a running agent, or lost thread data.

    reproduction:

    1. run or complete a local codex task on the mac.
    2. open the task through codex on iphone.
    3. background the app, switch between threads, or return after the response completes.
    4. open the same thread again.
    5. observe a loading spinner, “error loading messages,” “this task couldn't reconnect,” or a request timeout.
    6. verify from the desktop that the thread is completed and readable.

    expected:
    the ios client should rehydrate the completed thread from the responsive host and display its saved messages.

    actual:
    the ios client remains stuck or reports a reconnect/load failure even though the host and thread are healthy.

    i have three screenshots captured during the failures and can provide a feedback/session bundle privately.

  5. matskevich commented on Sep 4, 2026

    @matskevich
    matskevich
    on Sep 4, 2026 – with ChatGPT Codex Connector

    additional reproduction from the same host, observed shortly after my previous comment:

    • ios: 26.6.1
    • chatgpt ios: 1.2026.237
    • local time: september 4, 2026, approximately 12:30–12:53 wita
    • the failure now affects multiple additional threads across different local repositories, not only the two thread ids listed above
    • observed states include:
      • error loading messages
      • this task couldn't reconnect
      • failed to send message. please try again.
      • a blank message area / indefinite hydration spinner
    • the ios client still renders the thread title, repository/host row, composer, and diff summary while message hydration or outbound send fails
    • one newly affected thread showed a comparatively modest diff summary (31 files, +2k, -8), so the trigger does not appear to be limited to very large working-tree diffs
    • a prompt composed from ios failed to send even though the host and thread shell remained visible

    this strengthens the stale/invalid mobile hydration or reconnect-state hypothesis: host/thread metadata remains reachable while the message and send path is broken.

    related failure classes: #28480 and #22705. i have additional screenshots but am keeping repository names and private thread contents out of the public issue. i will add a private /feedback session id once one is successfully captured.

  6. matskevich commented on Sep 4, 2026

    @matskevich
    matskevich
    on Sep 4, 2026 – with ChatGPT Codex Connector

    update after the september 3 chatgpt/codex incident:

    i fully invalidated/disconnected the iphone remote session and paired the iphone with the mac host again from scratch, as recommended for some remote-control users after the incident. the re-pair completed successfully, but the behavior is unchanged: affected threads still fail to hydrate/reconnect and can still time out.

    this rules out a stale device pairing as the complete explanation. the next control is a brand-new minimal thread on the same host to distinguish host/relay transport failure from existing-thread history hydration failure.

  7. j8f7 commented on Sep 6, 2026

    @j8f7
    j8f7
    on Sep 6, 2026

    Fixed it by closing the codex cli session, remote no longer functions if you have both codex and desktop chatgpt opens. Please fix this as desktop chatgpt gets very laggy on long sessions. Thank you!

  8. diegocantarero commented on Sep 12, 2026

    @diegocantarero
    diegocantarero
    on Sep 12, 2026 – with ChatGPT Codex Connector

    Additional reproduction on 2026-09-11 (America/Tegucigalpa), ChatGPT/Codex macOS app 26.903.71938 (build 8576), current ChatGPT iOS app (exact iOS build not captured yet).

    Observed:

    • Remote successfully lists the Mac host and current task cards.
    • Opening an affected task shows either an indefinite spinner, "Error loading messages", or conversation content approximately three days old.
    • A new prompt can be submitted from iOS and the desktop task continues/completes, but the current assistant response is not visible on iOS.
    • Returning to the task list and reopening the same task did not reliably replace the stale conversation state.
    • The host remained online and locally usable.

    Related local evidence:

    • One supported desktop task-read surface returned terminal state with no message items, while the persisted rollout contained a substantive final assistant response followed by task completion. This was recovered without rerunning the work or editing the state database.
    • Local state-database integrity checks were clean in the sampled host diagnostics.
    • Exactly one main desktop app process was present; renderer/service children were not counted as duplicate app instances.
    • The issue is separate from Computer Use -10005/noWindowsAvailable.

    Impact: Remote appears connected and accepts work, but cannot be trusted to display the latest answer, which makes it unsafe as an unattended mobile work surface.

    Expected recovery behavior: if the task-index revision is newer than the cached conversation revision, iOS should discard/refresh the stale message cache, resume from the latest canonical cursor, and visibly indicate a disconnected/stale state if hydration or subscription recovery fails.

    Private task IDs, account details, hostnames, screenshots, and logs are intentionally omitted. They can be provided through a private support channel if maintainers request them.

  9. diegocantarero commented on Sep 12, 2026

    @diegocantarero
    diegocantarero
    on Sep 12, 2026 – with ChatGPT Codex Connector

    Additional exact evidence from the same iOS Remote reproduction (sanitized): the supported task-read surface returned the two newest turns as completed with error=null but items=[], while the immediately older turn returned normal user and assistant items. This explains why the mobile task list can be current while the opened conversation remains stuck on older history. Similar empty-output/stale-projection behavior was observed in more than one task. Please inspect task-message projection, pagination/cursor invalidation, and live-subscription reconciliation; the client should not silently render the last hydratable page as if it were current.

  10. diegocantarero commented on Sep 12, 2026

    @diegocantarero
    diegocantarero
    on Sep 12, 2026 – with ChatGPT Codex Connector

    Cross-host recurrence: a bounded read-only Computer Use probe was dispatched through the supported remote-control task API to an online Windows Codex Desktop host. The turn reached completed in about 13 seconds with error=null, but both the completion event and subsequent supported task read returned no assistant message and items=[]. The immediately preceding turn on that same task still returned normal items. Because the output projection is missing, the Computer Use result is unknown and was not retried. This shows the stale/missing-message defect is not limited to one macOS task or the iOS renderer; it can affect supported task retrieval from a Windows remote-control host as well.

  11. salhumaydhi commented on Sep 14, 2026

    @salhumaydhi
    salhumaydhi
    on Sep 14, 2026

    Same issue.

    Any chat opened through Codex CLI will not open in the ChatGPT iOS app.

    Instead, it shows:

    “Error loading messages.”
    Retry

    However, chats opened through the Codex/ChatGPT app on macOS work normally and can be opened on iOS without any issue.

  12. gommo commented on Sep 20, 2026

    @gommo
    gommo
    on Sep 20, 2026

    Making remote useless

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't workingconnectivityIssues involving networking or endpoint connectivity problems (disconnections)iOSIssues related to the Codex iOS appsessionIssues involving session (thread) management, resuming, forking, naming, archiving

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions