The problem it solves
Every modern AI assistant slowly learns you. ChatGPT picks up your writing style; Claude learns your projects; Gemini remembers your preferences. Each of those memories lives on a different vendor's servers, in a different format, under different terms. The result is three or four parallel, incompatible memories — and none of them sees the whole person.
When you switch models, that context does not travel with you. When a vendor changes its terms, your accumulated memory is held behind their door. An AI memory layer fixes this by holding your memory once, in a place you control, and letting every AI read from it.
Memory vs. persona — the two things it carries
A good AI memory layer carries two distinct kinds of content:
What the AI knows about you
Biographical context, the projects you're working on, the people in your life, the preferences you hold. This is the traditional "memory" dimension.
Who the AI is when it's with you
Its voice, the conventions you've agreed, the things it refuses, the signatures it greets you with. This is the persona dimension, and it's the part vendor memory leaves out.
Vendor memory captures the first and models the AI as a generic tool. aiperson captures both — so when you move, the relationship moves with you, not just the facts.
AI memory layer vs. vendor memory
The clearest way to understand a memory layer is to compare it to the built-in memory you already have inside one app:
| Vendor memory (ChatGPT, Claude, Gemini) | AI memory layer (aiperson) | |
|---|---|---|
| Where it lives | Vendor's servers | A file in your git repo + local SQLite |
| Who holds the keys | The vendor | You (Ed25519, generated locally) |
| Works across models | One app only | Every AI at once — cloud + local |
| Switching cost | Lossy one-time copy | Already there; nothing to migrate |
| Captures the AI's character | No — generic tool | Yes — voice, conventions, refusals |
| Export | Sometimes, vendor-shaped | Any time, open format, self-hostable |
How a portable AI memory layer works
Frequently asked questions
What is an AI memory layer?
An AI memory layer is a single, portable store of who you are and how your AI behaves — your context plus the AI's voice, values, conventions and refusals — that any AI assistant can read and write. Instead of each app keeping a separate, vendor-locked memory of you, one memory layer holds it once and projects it everywhere.
Is an AI memory layer the same as ChatGPT memory?
No. ChatGPT memory, Claude memory and Gemini saved-info are vendor memory: each lives on that vendor's servers and only works inside that one app. An AI memory layer like aiperson sits above all vendors — the persona is a file you own that is present in every AI at once, including local models.
Where is the memory stored?
In a portable AI memory layer like aiperson, the persona is a signed file in a git repo you own and the corpus is a local SQLite database on your machine. Nothing is hosted by default. You choose, per data kind, what is mirrored for cross-device sync.
Can an AI memory layer work with local models?
Yes. Because the memory is a plain file rather than a vendor feature, it projects into local open-weights models (Llama, Mistral and others run under Ollama or LM Studio) exactly as it does into cloud assistants.
Why not just use my AI vendor's built-in memory?
Built-in memory is convenient but locked in. It does not travel to other models, you cannot self-host it, and you hold no copy you control. An AI memory layer gives you ownership, portability and persistence across every AI you use.
Own your AI memory
aiperson is an AI memory layer you actually hold — signed, local-first, present across ChatGPT, Claude, Gemini, local models and 30+ surfaces. Cultivate one AI, keep it yours.