Can ChatGPT and Claude share the same memory?

Give it its due

Native memory is good, and it is doing a different job

ChatGPT keeps saved memories that the model updates as you work, alongside custom instructions and an option to reference past chats. You can see the list in settings, add to it, and delete from it. Claude's memory is on by default on Free, Pro, and Max plans, and off by default for Team and Enterprise until an admin turns it on. If you use projects, Claude creates a separate memory for each project so client work stays apart from everything else, and everything it remembers appears as a list of files under Topics that you can read, edit, or delete. It carries across chat and Cowork.

For personal working style this is genuinely useful. It learns that you want short replies, that you write in British English, that you are the person who always wants the numbers before the narrative. None of that needs a separate product, and HexaHQ is not trying to replace it.

Sources: OpenAI Help Center, "Memory FAQ"; Anthropic, "Bringing memory to teams"; Claude Help Center, "Importing and exporting your memory"; Claude Help Center, "Use Claude's chat search and memory". All checked 28 August 2026.

Where it starts to break

It is one person's memory, in one vendor's product

It belongs to a person, not to the company

Everything either assistant has learned about how your company works sits in one employee's account. On Team and Enterprise plans an owner turns memory on or off for the organization, which is a control over availability and not a way to share what one person's AI figured out. A new hire starts empty. When someone leaves, what their AI knew leaves with them.

It belongs to a vendor

Claude documents a way to import memory text from another assistant, and describes the imports as experimental, noting that Claude may not always successfully incorporate them. That is an honest description of what a snapshot transfer is. It moves personal preferences once. It does not keep two assistants in step, and it was never going to carry your renewal process.

Inside one vendor it can be split again

Claude's per-project memories are separate on purpose, which is the right call for confidentiality and the wrong shape for a company rule that should apply to every project at once.

There is no audience in between

Native memory has one audience: the person whose account holds it. There is no way to say that this pricing exception should reach the renewals team and not the contractors, or that this positioning language is approved for everyone while your draft notes stay private. Admin and project controls decide whether memory is available and where it is kept separate. They do not decide who a particular rule is for.

It remembers a person, not a process

"Writes in short paragraphs" is a preference. "Run the renewal 90 days out for this account because of a clause in their master agreement, and route anything above $1,000 in credit to Finance" is a process, with conditions, an owner, and consequences. Personal memory was not built to hold that, and it has no version history for it.

Knowing is not doing

Even a perfect memory stops at the answer. Updating the CRM, sending the notice, and issuing the credit means acting in real systems through the right person's account, with a policy on which actions need approval and a record of what happened.

Side by side

Three kinds of memory

	ChatGPT memory	Claude memory	HexaHQ
Whose it is	One person's account	One person's account	The organization, with personal context alongside it
Scope	Across your chats	Per project, across chat and Cowork	Per concept, resolved for the task at hand
Who else can receive it	Nobody, memory is per account	Nobody, memory is per account	The audience you choose, internal or external
Works with the other vendor	No	A one-time import, documented as experimental	The same artifacts across supported AI clients
Version history and corrections	Edit or delete entries	Read, edit, or delete entries	Versioned, with who corrected it and what the merge decided
Says when a rule applies to a deal	No	Only by which project you are in	Applicability recorded with the rule
Acts in your business systems	Through that client's own connectors	Through that client's own connectors	Any app with an API, per-person identity, allow / ask / deny

The other approach

Keep company memory with the company

Native memory is useful because it makes one AI product more personal over time. The limitation is ownership and portability. Company context that matters across people and tools should not have to be rebuilt when the organization changes AI clients.

HexaHQ stores the durable facts, rules, corrections, preferences, and ways of working that your organization chooses to keep. Supported AI clients can use the relevant parts without HexaHQ pretending to copy private model state or arbitrary chat history.

Be precise about what that does and does not mean, because the honest version is the useful one. HexaHQ does not copy transcripts between vendors. It does not read or move hidden model state. It does not sync your provider data from one assistant to another.

What it holds is what somebody deliberately saved: knowledge documents, reusable workflows, capability definitions, and the rules about where each applies and who receives it. Those are the things a company actually accumulates, and they are the things that should survive a change of vendor. ChatGPT, Codex, Claude Chat, Claude Code, and Claude Cowork all read them, so the sender can be working in one and the recipient in another and both start from the same organizational state.

Who gets what

One company context is not one company-wide memory

This is the point people most often get backwards. Moving durable context out of a personal memory does not mean pouring it into a shared pool that every employee and every AI can read. That would be a worse outcome than the problem it fixes.

Every piece of context in HexaHQ has an audience. Some of it stays personal, which is the right home for how you like your own drafts written. Some reaches one team or project group. Some is approved for the whole organization. Some is packaged for a client, a contractor, or an advisor outside the company. The person's AI inherits that boundary: it can use what its person is meant to have, plus that person's own working context, and it does not quietly gain the rest.

The company decides which people, and which AI acting for them, receive which pieces. That decision is separate from who wrote the document, and it can change later without duplicating anything.

A concrete example

The new hire's first week

An account executive spends eight months teaching her assistant how the company sells: which competitor claims to rebut and which to ignore, what a qualified opportunity looks like, the three things Legal always asks for before a security review, and the exact tone the CFO wants in a renewal note. By month nine her AI is genuinely good at her job.

A new rep joins. His assistant knows none of it. She can send him a document, which is a copy that will go stale, or he can spend eight months learning it the same way she did. If he prefers a different AI client to hers, even a memory export would not help much: it is a one-time text snapshot of her personal preferences, not the company's rules.

With HexaHQ, the parts of that which are company rules are documents shared with the sales team, and his AI reads them from his first task, in whichever supported client he uses. Her private notes on individual accounts stay hers. When Legal changes the security-review requirement, it changes in one place and both of their assistants pick up the new version. And when he asks his AI to send the renewal note, it goes from his own account, under the policy that governs that action.