Desktop performance collapses in profiles with a few very large local conversation histories (typing, scrolling, thread list, random exits)

[twentyOne2x]

Summary:
This is broader than slow thread switching. In profiles that contain a small number of extremely large local conversation histories, Codex desktop becomes slow across multiple surfaces: typing, scrolling inside a conversation, scrolling the thread list, delayed UI updates, slow thread switching, and occasional app exits. My strongest verified signal is not total conversation count, but the presence of a few giant local histories. Some of those heavy histories also had recurring thread heartbeat automations attached, which may amplify the problem, but that variable is not yet fully isolated.

Environment:

• Codex desktop app 26.415.40636
• macOS 26.3
• MacBook Pro (Apple M3 Max, 16 CPU cores, 128 GB memory)

Observed

• Typing can become very slow, roughly ~1s for words to appear.
• Scrolling inside an existing conversation can become very slow.
• Scrolling the thread list can become very slow.
• UI updates such as archive notifications can arrive with noticeable delay.
• Switching into some older conversations can take a very long time.
• The app can also exit unexpectedly.
• This appears Codex-specific on my machine: the slowdown can show up immediately inside Codex while the rest of the machine feels normal.

Verified A/B Repro

I tested this by comparing copied local profiles rather than only changing one profile in place.

Verified results:

• A fresh test profile was smooth.
• A copied profile that kept most of my original conversations but excluded the largest bloated ones was also smooth.
• A copied profile that restored only a few very large historical conversations became slow again, even without the rest of my normal active conversations.

What this establishes:

• The strongest correlation I can verify so far is the presence of a few giant local histories, not raw total conversation count by itself.
• This is broader than thread-switch latency. The same profiles also show typing lag, scrolling lag, delayed UI updates, and occasional app exits.

What I have not isolated yet:

• Whether simply archiving those same conversations inside the main profile would be enough to fix it.
• Whether recurring thread heartbeat automations targeting those same heavy conversations are contributing independently, or mainly amplifying an existing history-size problem.

Hypotheses

These are hypotheses, not claims of proof:

• The desktop app may be over-hydrating, resuming, or indexing a few massive local conversation histories.
• Stale stored workspace roots may be adding background churn during startup and/or resume.
• Recurring thread heartbeat automations attached to already-large conversations may amplify the problem by repeatedly waking expensive local histories.

My current read is that the issue is likely caused by some combination of:

1. oversized local history payloads
2. resume/read/index churn on those histories
3. stale workspace-root scanning
4. possibly heartbeat-driven repeated wakeups on the same heavy conversations

Diagnostics

Largest local histories on this machine, anonymized:

Local history	Rollout size	Turn contexts	Function/tool calls
#1	1867.27 MiB	618	21,369
#2	584.72 MiB	5,554	22,434
#3	561.02 MiB	554	22,831
#4	507.22 MiB	428	17,991
#5	455.84 MiB	563	38,073

Additional local signals:

• local state currently has 1767 total conversations, 777 unarchived
• 40 stored workspace roots no longer exist
• the desktop backend had 8 old session histories open totaling about 1.66 GiB
• desktop logs showed thread resume/read/unsubscribe operations taking tens of seconds, with worst cases around 82s
• startup also repeatedly scanned stale worktree/origin roots with very large expanded counts

Related issues

This may overlap with:

• #11011
• CLI-6413
• CLI-6450

[github-actions]

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

• Long Chat = Extreme Unresponsive. #18568
• Codex Desktop repeatedly starts full MCP stacks for new sessions/subagents, causing severe slowdown and memory pressure #18333
• Fix: Desktop App - Message send is delayed for ~8 seconds in new sessions after the latest update #18264

Powered by Codex Action

[venkatd]

I believe I am having the same issue. I have about 2-3 long running sessions (1 week long sessions). Switching sessions is slow, focusing on the chat box is slow. Things will lock up in all cases.

It's preventing me from enjoying it as much as I used to. Number one feature request is to speed it up. Thanks!

Shot in the dark, but does Codex do any virtualization in the chats? If not, perhaps using something like https://github.com/chenglou/pretext would unlock efficient virtualization by allowing calculating text heights efficiently.

[venkatd]

If it helps, I can confirm that archiving the conversations fixes the performance. Once I archived the long sessions and started new ones, things got much faster again.

One of my conversations was a bit over 1GB.

[wousp112]

I’m also seeing this multi-thread / large-history slowdown on Codex Desktop for macOS.

Setup:

• Mac with 16GB RAM
• Codex Desktop version: 26.417.41555 (build 1858)
• macOS version: macOS 26.4.1 (25E253)

Local state:

• ~/.codex/sessions is around 5.2GB
• Largest rollout file is around 207MB
• Several rollout files are over 80-130MB

Behavior:

• Disabling Memory made the app much smoother for single-window usage.
• However, opening multiple threads/windows still causes noticeable lag.
• Renderer processes continue to consume significant CPU/RAM when several threads are open.
• Thread switching and general UI responsiveness degrade as more large histories are present.

This seems related to large local history hydration/thread switching, possibly made worse when Memory is enabled.