Ask a friend who has known you for ten years for advice and they'll draw on everything they know about you: your job, your family, the promise you made last spring, the insurance claim that went wrong. Ask most AI assistants the same question and they start from almost nothing. Each new chat is a blank page.
That gap — between how much a human brain remembers and how little an AI assistant knows about you — is one of the biggest limits on how useful AI can be. It's also about to close. The question is who will own the memory when it does.
How human memory works (and why AI needs something similar)
Psychologists often divide long-term memory into two broad kinds. Episodic memory holds events: what happened, when, with whom. Semantic memory holds facts and meaning: your passport number, the name of your accountant, the fact that your car insurance renews in March. Together they let you act in context — to answer a question without having to re-learn your own life every morning.
Large language models have something like semantic memory about the world, learned during training. What they lack is durable memory about your world. Within a conversation they can hold a lot of context; when the conversation ends, most of it is gone.
AI is starting to remember — on the vendor's terms
The big assistants know this, and memory has become a headline feature. OpenAI has expanded ChatGPT's memory so it can refer back to past conversations, and Anthropic has rolled out memory for Claude. These are genuinely useful steps.
But they share a design choice: the memory lives inside one company's product. That raises practical questions people are only starting to ask:
- Is it portable? If you switch assistants, does your memory come with you?
- Is it complete? Your important information lives in PDFs, emails, photos and contracts — not in chat history.
- Who can read it? Is memory encrypted so that only you can read it, or is it readable by the provider?
- Can you see and control it? Can you check exactly what was remembered, audit who used it, and delete it?
Researchers studying how people perceive ChatGPT's memory have found that users are often unsure what is being kept and how it relates to their privacy — a sign that memory has arrived faster than the controls around it.
From chat memory to a personal memory layer
We think the right model looks less like a feature inside one chatbot and more like a personal memory layer: a single, private place that holds the things that matter in your life and your work, which any assistant you trust can draw on — with your permission.
That layer needs three properties:
- It's yours. You own the data and the keys. It isn't locked to one AI vendor.
- It's private by architecture. Content is encrypted so that even the company running it can't read it.
- It's permissioned. Each assistant or agent gets only the slice of memory it needs, for as long as it needs it, and every access is logged.
Why this matters now
AI is moving from answering questions to doing things: booking travel, filing paperwork, chasing invoices, renewing policies. Those tasks are impossible without context — and the context is your private data. As we explain in AI agents will need your private data, the more capable agents become, the more important it is that the memory they draw on is protected, portable and controlled by you.
Where AI Memory Vault fits
AI Memory Vault is our attempt to build that memory layer. You store documents, photos, notes and records once; the vault organises and connects them; you can ask questions in plain language, get reminded before things expire, and grant AI agents scoped, time-limited, logged access. The content is encrypted with AES-256-GCM before it leaves your device, with keys derived from your passkey. You can read more about how we're building it in Inside AI Memory Vault.
Humans forget. AI forgets. With the right memory layer — one you own — nothing has to.
Frequently asked questions
Do AI assistants remember past conversations?
Some now do. ChatGPT and Claude have memory features, but that memory usually lives inside one company's product and covers chat history rather than your documents and records.
What is a personal memory layer for AI?
A private store of your important information that any assistant you trust can draw on with your permission — owned by you, encrypted, and portable across AI tools.
Sources
- University of Georgia Grady College — “AI Never Forgets: Exploring Users’ Perceptions of AI Memory and ChatGPT’s Memory Feature” (CHI 2026)
- Vellum — Best personal AI assistants with memory (2026 review)
Figures are as reported by the sources above at the time of writing; survey data from vendors is indicative. AI Memory Vault is in private beta — product details describe how it is designed.