"Memories" consuming rate limit disproportionately

Resolved 💬 4 comments Opened Apr 23, 2026 by amanzoori-hm Closed Apr 23, 2026

What version of the Codex App are you using (From “About Codex” dialog)?

26.417.41555

What subscription do you have?

Enterprise

What platform is your computer?

Microsoft Windows NT 10.0.19045.0 x64

What issue are you seeing?

Since I activated "memories", my weekly rate limit has been going down disproportionately when Codex is idle. In the very beginning, it ate up almost the entirety of the remaining 40% of my weekly limit while the app was idle. I thought it was normal, as it had to go through all the existing threads. But then it continued. Now whenever I leave Codex idle, even if the only "new thread" since it last processed memories was a short one (< 500 tokens) during which I didn't do anything complicated, suddenly my weekly rate limit drops by 10% or more. Not sure if it's an issue with how inefficiently Codex is consuming tokens to generate memories, or a problem in how the consumption is calculated while memories are being created.

What steps can reproduce the bug?

  1. Activate memories in the Codex app on Windows.
  2. Wait for a few days until leaving the app idle doesn't cause token consumption anymore.
  3. Open a new thread, exchange a couple of messages. Check the remaining weekly rate limit at the end.
  4. Close any other Codex surfaces other than the app, and make sure all background/cloud tasks and automations are stopped.
  5. Leave the Codex app idle for ~1 hour.
  6. Check the remaining weekly rate limit again.

What is the expected behavior?

Based on normal rate limit consumption when actively using Codex, the weekly rate limit should be affected by no more than 2-3% for processing a short chat of ~400 tokens and extracting insights/memories from it.

Additional information

I just ended up disabling the memories feature because it renders Codex unusable with the current rate limit consumption.

View original on GitHub ↗

This issue has 4 comments on GitHub. Read the full discussion on GitHub ↗