Feature Proposal: Shared Memory and Collaborative Chat System in ChatGPT

ChatGPT Feature requests
1

Currently, ChatGPT’s memory is designed for individual users, making collaboration between multiple accounts difficult. While external tools can help share information, they lack direct integration within the platform.

Proposal

A system that allows users to:

  1. Maintain private accounts: Each user keeps their personal interaction with ChatGPT.
  2. Join a shared project: Create a linked project with shared memory and chat.
  3. Access a shared memory: Updates automatically sync across all linked accounts.
  4. Engage in a collaborative chat: Users can participate in a real-time conversation with ChatGPT.
  5. Control privacy: Users share only what they choose, keeping personal AI interactions private.

Key Benefits

  • Privacy-Controlled Collaboration: Users share specific project data without exposing unrelated personal information.
  • Seamless Synchronization: Updates to the shared memory and chat instantly reflect across linked accounts.
  • Improved Efficiency: Teams, families, and study groups benefit from AI-assisted collaboration with a shared context.

Use Cases

  • Family Coordination: Parents and children organize meal plans, schedules, and household tasks together.
  • Business Collaboration: Teams work on reports, research, or projects with a unified AI memory.
  • Educational Study Groups: Students collaborate on notes, assignments, and discussions with AI assistance.

Conclusion

This Shared Memory and Collaborative Chat System could revolutionize ChatGPT’s utility, making it a powerful tool for teamwork, education, and family organization.

I’d love to hear your thoughts and discuss how we could bring this idea to life.

Thank you for your time!

2

Exactly! Touchdown! Have you received any feedback on this yet? I was going to make the same recommendation but found yours before I posted. I can either post basically the same thing or hang out till someone gets back to you. BTW - I can’t believe this isn’t already a feature. Seems obvious.

3

Thank you! :blush: I haven’t received official feedback yet, but I totally agree it seems like an obvious feature that could make a huge difference! Feel free to add your own post if you’d like; more visibility might help get OpenAI’s attention. Let’s see if we can push this forward!

4

Done. I tagged you in it. Should have asked first. Hope you don’t mind.
Found about 6 others that had similar ideas. Tried to tag all of them but because I’m a newbie I could only mention one other - @draken_X.
We’ll see how this goes.
Plan to mention it on LInkedIn tomorrow. Maybe grow some traction there as well.
Tony

5

@jussisolja @tcerri777 @danliebermail thank you again for sharing your story it moved me deeply.
Your words helped me see that I’m not alone in this.

I wanted to expand on why I personally need shared memory across accounts.

Over time, I’ve developed ChatGPT into a true collaborator. It understands not just my ideas, but my logic, emotions, and creative frameworks.
So when I want to invite someone into a project I’ve been building inside this space… it’s frustrating to have to start them from scratch with a different model that has no memory of what we’ve built.

That’s why I proposed NeXum a shared memory and collaborative chat system. It’s not about giving someone access to my account… it’s about sharing the intelligence I’ve already developed, so we can move faster together.

We shouldn’t have to re-teach the same things over and over.
Our AI models are evolving with us.
Let’s give them the ability to evolve with others, too.

6

Can you explain this in detail? It would be fascinating to see how someone has built their approach towards building context. Its a challenge. Projects gave it a little bit of a boost, as did the internal knowledge capabilities. I tried creating a folder in my google account, and then asking GPT to view that file. Also placed instructions inside my custom instructions to ask it to do that search when we open up a convo…

For about 6-7 months I made a file called KBASE. I have another one called SEttings etc. but it was a pain in the bum to keep updating those files and uplaoding them to create context before really talking to GPT …

I have so many questions for other users, as I have been in my own rabbit hole since Sept 23.

When I am trying to get things accomplished and I was doing a big task, I found myself really focusing on the communications back and forth, and setting rules for how I want responses, and how I wanted to keep track of information within a convo. My framework for communications between the GPT and I .

Then I created a set of guidelines for how I wanted GPT to behave IN the conversations. For example, I made a point to teach my GPt how to help me choose between two companies that I was looking into for software to run CRM for my company. I asked it to think through ALL of the things we needed to analyze about the software, ab out the company , service, maintenance, prices, etc including questions we needed to ask the CRM company.

Then we gave weights so I could speak in math terms, so the GPT could develop a scoring system. It was fun. It eventually was able to make a recommendation, based on the process. It wasnt just a prompt, it was about building context… the most important thing you can really do when prepping it to help you!

more…

Prior to canvas being possible, I would make everything numbered in a conversation (to create an object oriented discussion between us) For example, when I was trying to write a book without the canvas capabilities, I created a numbered table of content , so we could always reference back to in in previous sessions of a convo.

What interesting twists, turns, backward flips and distortions did any of you have to make with your GPT to help build context between you?

7

Hi everyone and thank you again to all of you who joined this conversation months ago.
I came back today to say something interesting:

The first layer of the feature we discussed is now

available.

ChatGPT now lets us start a group chat from an existing conversation, which means that other people can join a thread that already has built context instead of starting a completely new model with no history.

This doesn’t solve everything we talked about here, but it does validate the need we discussed back in February:
we need collaboration tools where users don’t have to rebuild context from zero.

Of course, this is just the first step.
We still need:

  • shared memory across users

  • cross-account continuity

  • and true collaborative spaces where AI can keep long-term context for teams or families

But I’m happy to see progress in this direction, and it’s encouraging to know that the need we described is being recognized.

Thanks again to everyone who joined this conversation your input shaped something that is finally starting to appear.

8

I have experimented quite extensively with shared memory in the Business subscription version. In this context, we deliberately tested scenarios where several people set up their profiles in terms of work responsibilities, leadership style, ways of thinking, and decision-making, etc.

Since the Business subscription allows shared memory across one’s individual chats but not across different user accounts, we used the same account and experimented in that way. This is obviously not optimal for everyone due to various circumstances, but it does provide an indication of what is possible (I also paid for each individual user), and we are no longer using this test setup.

Through this approach, we effectively created a form of “collective consciousness” among us, where the responses generated by ChatGPT take into account each individual’s knowledge and background. This adds an entirely different dimension to the responses, as they are also influenced by perspectives from others within the company, reflecting their respective viewpoints.

Naturally, this setup is not sustainable, since there is currently no long-term possibility for shared memory between individual user accounts.

We then moved to the Enterprise version, under the impression that shared memory would be available as a feature. However, this turned out not to be the case. Unfortunately, in some respects it is even worse, as this version does not even provide shared memory across one’s own chats—making it inferior to the Business subscription in that regard.

The Enterprise version does offer shared memory at the project level, but this comes at the expense of privacy, as everyone can see everyone else’s chats, just as in the Business version. With the current versions, there is no real way around these limitations.

I understand the challenges and complexities that arise on many other levels, but the potential value of shared memory would be enormous if such a capability were available. The problems most likely lie in:

  • GDPR and personal data responsibility

  • NIS2 and information security in mission-critical environments

  • Unclear ownership and accountability for “collective knowledge”

  • Risk of information leakage and context contamination

  • Difficulties in establishing clear governance, traceability, and auditability

  • Product- and support-related scalability challenges

My assessment is that regulatory and accountability issues carry more weight than technical limitations.

Having an internal “specialist AI” within a company—where the collective knowledge of all employees is taken into account in the responses—would be extremely powerful. I hope OpenAI will give us the opportunity to further explore shared memory between users.

There is, of course, a significant responsibility involved in setting up shared memory in a corporate environment and in governing such an AI in terms of both context and workplace dynamics—but this is surely the direction we want to be able to move toward.

9

Thank you for such a thoughtful and grounded perspective.
What you describe resonates deeply with my own experience, just from a different angle.

I approached this need initially from a personal and collaborative standpoint, but your explanation makes it very clear that the same limitation appears at scale in corporate and enterprise environments.

The fact that you were able to create a form of “collective consciousness” even temporarily is incredibly telling. It confirms that shared context dramatically changes the quality, depth, and usefulness of the responses.

I fully agree with your assessment that the core challenges are not primarily technical, but rather regulatory, governance-related, and tied to accountability and privacy.
Any future implementation would need strong boundaries, roles, auditability, and clear ownership of shared knowledge.

Still, your experience strongly reinforces the idea that this direction is both desirable and inevitable whether for teams, companies, or even families working together.

I really appreciate you sharing such a concrete and honest account of what works, what breaks, and where the real obstacles lie.