›_ Codex Issues
Codex Issues / #26235

Codex Desktop hangs after sending prompt on macOS

Open 💬 3 comments Opened Jun 3, 2026 by bayshores
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What version of the Codex App are you using?

Codex Desktop 26.601.21317 (CFBundleVersion 3511)

Bundled CLI:

codex-cli 0.136.0-alpha.2

What subscription do you have?

ChatGPT account. I can use Codex CLI successfully on the same machine/account.

What platform is your computer?

ProductName:    macOS
ProductVersion: 26.2
BuildVersion:   25C56
Darwin:         25.2.0 arm64

Apple Silicon Mac, arm64.

What issue are you seeing?

Codex Desktop opens and the shell UI loads, but starting a new chat and sending even a tiny prompt hangs forever.

Observed behavior:

  • Sidebar loads.
  • Projects load.
  • Environment panel loads.
  • New chat input is usable.
  • After sending test, the app remains stuck loading and no assistant response appears.
  • The CLI works, so this does not appear to be a general local Codex auth or CLI failure.
  • Reinstalling the Codex app from the official macOS DMG did not fix it.
  • Launching with a clean temporary app profile did not fix it.

Steps to reproduce

  1. Open Codex Desktop on macOS.
  2. Select any local project.
  3. Click New chat.
  4. Send:
test
  1. Observe that the turn hangs indefinitely.

Expected behavior

Codex Desktop should start a turn and return a normal response for a trivial prompt like test.

Actual behavior

The UI stays in a loading state indefinitely. The user cannot get a response from Codex Desktop, even though the CLI works.

Troubleshooting already attempted

I tried the following without resolving the issue:

  • Confirmed bundled CLI version:
/Applications/Codex.app/Contents/Resources/codex --version
codex-cli 0.136.0-alpha.2
  • Checked app logs in:
~/Library/Logs/com.openai.codex/2026/06/03
  • Killed stale old codex app-server and node_repl helper processes.
  • Moved temporary bundled marketplace cache out of the way:
~/.codex/.tmp/bundled-marketplaces
  • Removed deprecated config warning:
[features].web_search_request
  • Pressed Settings > Reset and install Workspace. The UI said it could not reinstall dependencies, but logs reported the primary runtime as installed with problemCount=0.
  • Temporarily disabled external MCP servers in ~/.codex/config.toml, including context7, stripe, and shaders. This did not fix the hang. Config was restored afterward.
  • Reset Electron/browser cache folders under:
~/Library/Application Support/Codex
  • Launched Codex with a clean temporary app profile:
~/Library/Application Support/Codex-clean-test-20260603-1253

The clean profile also hung.

  • Reinstalled the Codex app from the official Apple Silicon DMG listed in the Codex docs:
https://persistent.oaistatic.com/codex-app-prod/Codex.dmg

The old app was moved aside and a fresh /Applications/Codex.app was installed. The issue still reproduces.

Relevant log snippets

Startup and CLI handshake succeed:

initialize_handshake_result durationMs=89 initializeRequestId=__codex_initialize__ outcome=success transportKind=stdio
Codex CLI initialized
app_server_connection.state_changed ... next=connected

Common repeated warnings/errors seen while the Desktop app hangs:

Statsig: error while bootstrapping post-login client ... path ["user", "custom", "workspace_type"] ... Invalid input
Timed out waiting for MCP response to marketplace/add
Timed out waiting for MCP response to experimentalFeature/list
Timed out waiting for MCP response to externalAgentConfig/detect

Dependency diagnosis says the primary runtime is installed:

primary_runtime_dependencies_diagnose_started hostId=local
primary_runtime_dependencies_diagnose_finished bundleVersion=26.601.10930 installed=true problemCount=0

After the hang, logs show normal app-server responses for many setup calls, but the user-visible turn never completes in the Desktop UI.

Similar existing reports

This looks related to these reports:

  • #24287, Codex Desktop accepts prompt but UI stays stuck in Thinking
  • #13715, macOS local Codex clients hang after sending prompts
  • #24467, Codex Desktop threads remain spinning after long turns
  • #11090, orphaned app-server processes and UI freezes

Additional notes

This does not look like a simple cache, profile, MCP, or app bundle corruption issue because the issue persists after:

  • clean temporary profile
  • external MCP disable test
  • Electron cache reset
  • fresh official app reinstall

The CLI is still usable on the same machine.

View original on GitHub ↗

3 Comments

github-actions[bot] contributor · 2 months ago

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

  • #26162
  • #25170
  • #24850

Powered by Codex Action

bayshores · 2 months ago

Clarification after reviewing the duplicate candidates flagged by the bot:

  • #26162 is Windows + WSL. This report is macOS arm64.
  • #25170 is also Windows + WSL. This report is macOS arm64.
  • #24850 is closer, but it focuses on Desktop remaining in Thinking after successful tool calls or a hung follow-up /responses request. This report reproduces on a new chat with the first trivial prompt test, before any visible tool calls or assistant output.

This issue also persisted after:

  • clean temporary Desktop profile
  • external MCP servers disabled
  • Electron cache reset
  • fresh official macOS app reinstall

The most distinctive local log signal so far is repeated Desktop/app-runtime timeout behavior plus the Statsig bootstrap error for user.custom.workspace_type, while the bundled CLI handshake succeeds.

AdiedX · 6 days ago

Thanks for documenting this. I hit a very similar macOS Desktop symptom: a trivial prompt appeared to submit, but no assistant turn started, while the bundled and standalone CLI were otherwise healthy. In my case the root cause was local hook configuration, not authentication, the model, or the backend.

My SessionStart hook was hard-coded to load a compatibility AGENTS.md source and required a section named Mandatory topic files. The canonical instruction file no longer contained that section. The hook therefore returned continue: false with an instruction-loading error. Codex exited successfully at the process level but emitted no agent message, which made the Desktop behavior look like a silent prompt hang.

The generic diagnostic and recovery path was:

  1. Inspect ~/.codex/hooks.json and every SessionStart hook script.
  2. Invoke the hook directly with a representative SessionStart payload.
  3. Inspect continue, stopReason, systemMessage, and hookSpecificOutput.additionalContext.
  4. Make sure the hook resolves the actual canonical AGENTS.md path instead of relying on a stale compatibility copy.
  5. Treat optional companion files, sections, and includes as optional. Do not fail the entire session because an optional instruction section is absent.
  6. Remove unsupported configuration fields from events that cannot emit them. In my case, additionalContextLimit on the PreCompact hook was generating another warning.
  7. Retest both the standalone and bundled Codex binaries with --ephemeral, before deleting session state.

After correcting the hook, 10/10 standalone probes and 10/10 ChatGPT-bundled Codex probes returned an actual OK agent message. This will not explain every Desktop hang, especially failures involving Remote Control, CryptoTokenKit, renderer CPU loops, or app-server IPC, but it is a concrete pre-turn failure mode worth checking before reinstalling the application or deleting local state.