What is ChatGPT Memory and how does it work?

ChatGPT Memory is a feature inside OpenAI's product that lets the assistant remember details across conversations — things like your name, work, projects you're on, and preferences you've stated. It's controlled from Settings → Personalization → Memory. OpenAI's models silently decide what's worth saving, you can review and delete entries, and the memory only applies inside ChatGPT — it doesn't travel to Claude, Gemini, or any other tool.

How is Konshus different from ChatGPT Memory?

ChatGPT Memory is provider-owned, scoped to one product, and resets when you switch tools or OpenAI changes its memory architecture. Konshus is a separate vault you own — it ingests ChatGPT exports, Claude exports, journals, voice, documents, and social archives, distills them into atoms (small facts with source + confidence), and lets you export a portable persona that works with any model. The wedge is portability and ownership, not chat quality.

Can I export ChatGPT Memory?

Indirectly. OpenAI's data export (Settings → Data Controls → Export data) includes your Memory entries inside the ZIP alongside conversations. But it's a raw dump — not a structured persona, not importable into another tool, and not refreshed automatically. You have to request a fresh export every time you want a current snapshot.

Is ChatGPT Memory private?

It lives on OpenAI's servers, governed by their privacy policy. OpenAI states they don't train on ChatGPT Memory entries for consumer accounts when you've opted out of data training, but the data sits on their infrastructure and is subject to their access controls and changes in policy. Konshus is per-artifact encrypted, has a private-mode flag for sensitive entries, supports hard-delete with audit, and never uses your data for model training under any circumstance.

Does Konshus work with ChatGPT?

Yes — Konshus is designed to complement ChatGPT, not replace it. You import your ChatGPT export into the vault, Konshus distills it into atoms, and then you export a Whisper or Briefing (a compact persona summary) that you paste into the top of a ChatGPT thread. The result: ChatGPT acts like it remembers you, even when its own memory was reset or wasn't enabled.

What happens to ChatGPT Memory when OpenAI changes its architecture?

It has happened multiple times since 2024 — memory has been re-architected, re-scoped, and at least once silently reset for some accounts during a rollout. There's no SLA on continuity. If your assistant's recall of who you are matters to your work, the only durable answer is to keep a copy of your context somewhere outside any single provider.

Do I need Konshus if I'm already happy with ChatGPT Memory?

Probably not today. The case for Konshus gets stronger the more you've invested in a given AI, the more you'd lose if the memory disappeared, and the more often you switch between models. If you're a casual user with two months of light usage, ChatGPT Memory alone is fine. If you've spent two years building a working relationship with an assistant, a separate backup is cheap insurance.

How much does Konshus cost compared to ChatGPT Plus?

Konshus starts at (Spark plan, billed annually at $49/year). ChatGPT Plus is $20/month. They're not substitutes — most Konshus members keep their ChatGPT subscription. Konshus is the persistence layer; ChatGPT is one of the models you talk to. Full pricing at /pricing.

A comparison · ~7 min read

Konshus vs ChatGPT Memory — What Each One Actually Does

ChatGPT Memory and Konshus get compared a lot, and the comparison is slightly off. They're not competitors at the same layer. ChatGPT Memory is a feature inside one assistant. Konshus is a separate vault that holds the version of you that any assistant could use. Here's what each actually does, where they overlap, and where they don't.

The short version

ChatGPT Memory is provider memory — OpenAI's product remembering things about you while you're inside OpenAI's product. When you leave (switch to Claude, switch to Gemini, lose access, get reset during a rollout) the memory stays behind.

Konshus is portable memory — a vault you own that ingests your conversations, journals, voice, and documents, distills them into structured atoms, and exports a persona you can paste into any model. When ChatGPT changes, your context doesn't.

Most Konshus members keep using ChatGPT. The two are stacked, not substituted.

Side-by-side

Dimension	ChatGPT Memory	Konshus
Where memory lives	OpenAI servers	Your vault — exportable, deletable, encrypted at rest
Which models can read it	ChatGPT only	Any — ChatGPT, Claude, Gemini, Grok, local models
What gets remembered	Whatever OpenAI's model decides is salient	Atoms you confirm, with source + confidence
Survives model deprecation	Sometimes — depends on the rollout	Yes — the vault is independent of any model
Export format	Raw entries inside the ChatGPT data export ZIP	Whisper, Briefing, Full Mirror, Crisis Handoff
Imports from other tools	No	ChatGPT, Claude, paste-text, journal, documents, social ZIPs
Private by design	Subject to OpenAI's policy	Per-artifact encryption, private-mode flag, never trained on
Cost	Included in ChatGPT (free or Plus)	From ~$4/month (Spark, billed annually)

What ChatGPT Memory is genuinely good at

Worth being clear: ChatGPT Memory is a real feature that does real work. It's frictionless — you don't have to do anything, the model quietly saves what it thinks matters. It's free or included in Plus. It makes conversations feel less repetitive after a few weeks. For someone using one tool casually, it's plenty.

The limits show up when any of three things happen: you start using another model, your account gets reset or suspended, or the memory architecture changes in a rollout and your assistant quietly stops sounding like the one you knew. All three have happened to enough people that "what's my backup plan?" is a fair question.

What Konshus does that ChatGPT Memory can't

Portability. The persona exports (Whisper, Briefing, Full Mirror) are designed to paste into any model's system prompt or first message. ChatGPT acts like it remembers you. Claude does too. So does Gemini. The memory follows the person, not the provider.

Structure. A ChatGPT Memory entry is a sentence. A Konshus atom is a sentence with a source (the conversation it came from), a confidence score, a timestamp, and a thread to your stated identity. You can audit where any claim came from and retire any one that's gone stale.

Ingestion from outside ChatGPT. Konshus reads your Claude export, your journals, your voice notes, your documents, and your Facebook/Instagram/LinkedIn archive ZIPs. Years of context that ChatGPT was never going to touch becomes part of the vault.

Survival. When OpenAI deprecates a model, your context is unaffected. When ChatGPT changes its memory rules, the vault doesn't notice. When you decide to leave a provider entirely, you walk out with your memory in hand.

Short answer

ChatGPT Memory and Konshus do different jobs. ChatGPT Memory helps ChatGPT remember you while you're inside ChatGPT. Konshus holds the version of you that any model — including ChatGPT — can read, owned and exportable by you, surviving model updates and provider changes. Use ChatGPT Memory for in-product convenience. Use Konshus when continuity, portability, and ownership actually matter.

When ChatGPT Memory is enough

You only use ChatGPT, and you don't see that changing.
You've been using it for less than a year.
You'd shrug if your assistant suddenly forgot you tomorrow.
Your usage is mostly task-based — drafting, summarizing, coding — not relationship-based.

When Konshus is the better answer

You've been investing in an AI for a year or more and the loss of that context would hurt.
You use multiple models, or expect to, and resent retraining each one from scratch.
You write, journal, or think out loud with AI in a way that has cumulative value.
You want the data on your terms — encrypted, exportable, never trained on.
You've already been burned once by a memory reset or a model deprecation.

The honest comparison

OpenAI built ChatGPT Memory to make ChatGPT stickier — which is the right business decision for OpenAI. It also means the memory is, structurally, theirs. If that arrangement works for you, you're done reading.

Konshus exists because some people would rather own the version of themselves that AI uses. Not as a hedge against OpenAI in particular — as a hedge against the simple fact that platforms change, models get deprecated, and the most valuable artifact of the AI era is the cumulative record of who you are. We think that record should belong to you. Related reading: Konshus vs Claude Projects, best AI memory systems, and what to do when ChatGPT memory is full.

Frequently Asked Questions

Your memory shouldn't depend on one company's roadmap.

Konshus ingests your ChatGPT and Claude exports, distills them into a structured vault, and gives you a portable persona that works with any model — encrypted, exportable, yours.

Meet Konshus