Feature Request: Persistent Context Pins — Give Users Control Over What ChatGPT Never Forgets
Product: ChatGPT
Category: Memory / Context Management / User Experience
Priority: High-value product enhancement
1. Overview
I would like to propose a new feature called Persistent Context Pins.
The idea is simple: users should be able to manually mark specific messages as important, ensuring that these messages remain protected from ordinary context compression and retain the highest priority when ChatGPT reconstructs context for future responses.
Today, AI systems can summarize or compress earlier conversations when context windows become full. Although this helps maintain conversation continuity, important details, instructions, preferences, and decisions may be lost or oversimplified.
Users should have more direct control over what AI remembers.
Instead of relying entirely on AI to decide which information matters, users should be able to explicitly identify information that must be preserved.
2. Proposed User Interface
Add a small star icon (
) to the message action toolbar, ideally next to the Copy button.
When a user clicks the star, the message becomes a protected context item.
The default action should:
-
Pin the original message in the current conversation.
-
Save the information to the user’s global ChatGPT Memory, when Memory is enabled.
-
Give the message the highest priority among ordinary user-provided conversation context.
-
Protect it from being silently removed or rewritten during context compression.
Users should also be able to manage the two scopes independently:
-
Conversation-only: Keep the information protected in the current conversation.
-
Global Memory: Make the information available across different conversations.
Ideally, both should be enabled by default when a user stars a message, with the ability to change these settings.
3. Two-Level Memory Architecture
I propose two interconnected memory layers.
A. Protected Conversation Context
This layer stores messages explicitly pinned by users.
Its requirements:
-
Original messages must be preserved.
-
Ordinary context compression must not delete pinned messages.
-
Pinned information must be prioritized whenever the conversation context is reconstructed.
-
Pinned messages should be loaded before ordinary conversation summaries.
-
The system must maintain their original meaning and, where requested, their exact wording.
Most importantly, pinned messages should remain accessible regardless of how long the conversation becomes.
B. Global Persistent Memory
When a user stars a message, the information can also be synchronized with their account-level Memory.
This allows important preferences, personal context, working instructions, and long-term decisions to remain available in future conversations.
The original message should remain linked to the saved memory so users can review its source.
4. Context Compression Protection
This is the most important technical requirement.
When ChatGPT approaches its context window limit, it may summarize or compress ordinary conversation history.
However, user-pinned messages should be treated differently.
I propose the following context reconstruction priority:
-
System and safety instructions.
-
User-pinned protected context.
-
Relevant global memories and project context.
-
Recent conversation history.
-
Compressed summaries of older conversations.
Pinned messages should receive a reserved context budget.
If the pinned content exceeds that budget, ChatGPT should notify the user and offer options to manage it. It should never silently remove protected content.
For users who select exact preservation, the original message must remain intact even if an additional condensed representation is used for context efficiency.
5. Additional Features
A. Memory Management Panel
Add a panel where users can view all their pinned messages.
It should display:
-
Original message.
-
Source conversation.
-
Whether it is synchronized with Global Memory.
-
Token usage.
-
Preservation mode.
-
Date pinned.
B. Preservation Modes
Offer two options:
-
Exact Preservation: Keep the original message unchanged.
-
Meaning Preservation: Allow ChatGPT to create a shorter representation while retaining the original source and its intended meaning.
C. Conflict Detection
If a newly pinned message contradicts an older memory, ChatGPT should notify the user and ask whether the new information should replace or coexist with the previous one.
For example:
Previous memory: “I work at Company A.”
New pinned message: “I have joined Company B.”
ChatGPT should identify the potential conflict instead of blindly using both.
D. User Control
Users should always be able to:
-
Unpin a message.
-
Remove it from Global Memory without necessarily unpinning it in the current conversation.
-
Edit or replace a pinned item.
-
Review its original source.
-
View previous versions.
6. Technical Feasibility
This feature does not require every conversation message to remain permanently loaded into the model’s active context.
Instead, pinned messages can be stored separately from ordinary conversation history.
A context assembly system can then prioritize these protected records and inject them into each relevant model interaction.
A reserved token budget would make the guarantee practical while preventing unlimited growth in context size.
The system should also distinguish between permanent storage and active context inclusion. Keeping an original message safely stored does not, by itself, guarantee that the model has read it during every interaction.
7. Why This Matters
This feature would solve several important user problems:
-
Users would no longer need to repeatedly explain important information.
-
Long conversations would become more reliable.
-
Important instructions would be less vulnerable to context compression.
-
Users would have more control over AI memory.
-
Long-term projects and relationships with AI would become more consistent.
-
Users could trust that explicitly protected information will not be silently forgotten.
I believe this is particularly valuable for people who use ChatGPT as a long-term personal assistant, business assistant, research partner, or project collaborator.
8. Suggested MVP
I recommend starting with five core capabilities:
-
Add a star icon to individual messages.
-
Create a protected context area for starred messages.
-
Prioritize these messages during context reconstruction.
-
Provide an option to synchronize starred messages with Global Memory.
-
Notify users when protected context approaches its reserved token budget.
More advanced features, such as conflict detection, version history, and different preservation modes, can be introduced later.
9. Expected Product Value
Persistent Context Pins would turn memory from something users largely depend on AI to manage into something users can actively control.
The central idea is:
AI should decide what is useful to remember automatically, but users should have the final say over what must never be forgotten.
I believe this would be a meaningful improvement to ChatGPT’s long-term usability, reliability, and user experience.
Thank you for considering this suggestion.