Codex Desktop on macOS was very slow today and kept reconnecting to the local app-server

Summary

Today Codex Desktop was very slow to use on macOS, and it also kept reconnecting to the local app-server.

The core problem I want to report is not just "general performance feels bad". The main symptom today was:

1. access/use felt slow
2. the desktop app repeatedly reconnected

I originally filed this too broadly, so this updated report is specifically about slow access together with repeated reconnect behavior.

Environment

• Codex Desktop
• macOS 15.7.4
• App-server version seen repeatedly in today's logs: 0.122.0-alpha.13
• Runtime bundle: 26.416.11627

What I observed

• Codex felt slow throughout today
• While using it, the app repeatedly appeared to disconnect and reconnect
• The issue felt like connection churn, not just a heavy thread rendering slowly once

Expected behavior

• Codex Desktop should stay responsive during normal use
• The local app-server connection should remain stable instead of repeatedly reconnecting

Actual behavior

• Access/use was slow today
• The app repeatedly went through reconnect-like connection cycles

Local evidence from today

For 2026-04-23:

• 29 codex-desktop-*.log files were generated
• Today's logs contain about 15 cause=start_process state transitions
• Today's logs contain about 15 cause=stop_process state transitions
• ~/.codex/logs_2.sqlite is about 753 MB
• ~/.codex/logs_2.sqlite currently contains about 151,652 log rows

Other signals that showed up while this was happening:

• Received turn/started for unknown conversation: about 159 occurrences
• Not a git repository: about 6,718 occurrences
• unsupported feature enablement 'realtime_conversation': about 138 occurrences
• 401 Unauthorized: about 2 occurrences

Example log snippets

Repeated connection start:

info [AppServerConnection] app_server_connection.state_changed cause=start_process ... next=connecting ... transport=stdio

Repeated connection stop:

info [AppServerConnection] app_server_connection.state_changed cause=stop_process ... next=disconnected ... transport=stdio

One stop event included this error context:

failed to refresh available models: timeout waiting for child process to exit

Another stop event included:

failed to warm featured plugin ids cache ... 401 Unauthorized

I also saw:

error [electron-message-handler] Received turn/started for unknown conversation conversationId=...

Notes

• The main reason I am filing this is the combination of slow access plus repeated reconnect behavior
• The extra log noise above may or may not be causal, but it was present during the same time window
• If useful, I can provide a narrower time window or more exact raw log excerpts around a reconnect event