It seems like OpenAI is on the verge of rolling out an improved ChatGPT memory feature. Obviously, the company has not revealed any details about it. But an X user named Tibor Blaho has revealed that we might see a new, improved ChatGPT memory feature soon.
According to Blaho, a new “Golden Hour” prompt has appeared in the ChatGPT web app, which hints towards a forthcoming feature called Automatic Memory Management.
I think we might see a new improved ChatGPT memory feature soon, because there's a new "Golden Hour" announcement in the web app
— Tibor Blaho (@btibor91) September 12, 2025
Saved memories had the codename "Sunshine" (save and use memories) and chat history had the codename "Moonshine" (reference previous conversations) pic.twitter.com/6oJwFjRCdM
He also suggests that earlier codename designations may hint at the feature’s architecture:
- Sunshine was reportedly the codename for “saved memories” (i.e., ChatGPT storing personal facts or preferences)
- Moonshine was used for chat history (i.e., referencing past conversations)
In Blaho’s view, the “Golden Hour” announcement may be the umbrella under which OpenAI launches “Automatic Memory Management,” a system that decides what to retain, what to discard, and when to surface memory only when it’s relevant. Basically, preventing memory tables from being “full.”
What could this mean for Us?
If this leak turns out to be true, then it could address one of the more frustrating limits in current ChatGPT usage: reaching memory or storage caps, which force us to trim or overwrite what the model “remembers.” With this upcoming update, we might not need to manually manage memory slots, and they would (in effect) trust the model to make those choices.
Apart from that, it might signal steps toward more nuanced memory policies: the system could prioritize recent, high-value, or frequently used data while aging out irrelevant or stale memories. In an ideal version, it might maintain “just enough” context to be helpful without drifting or overcommitting to past user inputs.
However, these are all just speculations for now, until OpenAI confirms or releases the feature. So while we should consider these leaks as an interesting hint, they are not a guarantee. I’ll keep tracking developments and share updates once more concrete evidence emerges.
