Per-Conversation Toggle for Memory and Chat History Context

ChatGPT Feature requests
1

Hi OpenAI Team,

I’d like to suggest a new feature for the ChatGPT desktop application regarding how memory and chat history are referenced in conversations.

:wrench: Feature Proposal

Currently, users can globally enable or disable whether ChatGPT references memory and past conversations via the settings. I propose an additional feature that allows:

  1. Default Behavior Based on App Settings: When starting a new conversation, the default state (memory and chat history on/off) should follow the user’s application settings.
  2. Per-Conversation Override Option: Before initiating a new conversation, users should be able to override the default state by choosing whether the model should ignore memory and chat history for that session. This allows starting a “fresh” session where the model won’t reference any prior context or user memory.
  3. In-Conversation Context Control (Optional Extension): During a conversation, users could optionally be given a toggle to decide if a particular assistant reply should consider previous messages and memory, or respond as if it were context-free.

:bullseye: Use Cases

  • Model Comparisons: To ensure fairness when comparing outputs across models like GPT-4, GPT-4.1, and GPT-4o, it’s important that no past context or memory biases the results.
  • Research/Testing Scenarios: For example, when asking ChatGPT to develop an iOS app from scratch, we don’t want it to “know” the user is an iOS developer based on prior interactions—this could skew the outcome of the experiment.

:thinking: Why Not Just Use Settings?

While users can disable memory or chat history in settings, this only works globally. The risks include:

  • Forgetting to re-enable memory after testing, causing typical conversations to lack context or miss important memory entries.
  • Forgetting to disable memory before testing, leading to biased results in research or model evaluation.

Having a per-conversation override offers flexibility, avoids unintentional memory usage, and preserves long-term settings.

Thank you for considering this feature! It would be incredibly helpful for researchers, testers, and users who require granular control over session behavior.

(This post was drafted with the help of ChatGPT, based on my own use case and feature requirements. With ChatGPT’s assistance, I was able to articulate this feature request more clearly and precisely.)