What Claude's memory does well
Anthropic's built-in memory lets Claude remember you across conversations — your writing style, the projects you're carrying, the way you like an answer shaped. For anyone living inside Claude it is a real improvement, and it is getting better with each release. We are not here to talk you out of it. We use Claude too.
The question this page answers is the one that surfaces the moment you use a second AI, or run a local model, or wonder what happens if the terms change: whose memory is it?
Where Claude's memory stops
- One app. Claude's memory works inside Claude. ChatGPT, Gemini and your local Llama see none of it.
- Vendor-held. It lives on Anthropic's servers, in Anthropic's format, under Anthropic's terms. You hold no copy you control.
- A one-time copy out. Since March 2026 you can import a memory when you switch in, but it is a lossy, point-in-time copy — not a living memory that keeps in sync. Use two AIs and you are back to two memories.
- Facts, not character. Vendor memory captures what the AI knows about you and models the AI as a generic tool. The voice you've cultivated, the conventions you've agreed, the things it refuses — the persona — is the part that doesn't travel.
Claude memory vs. an AI memory you own
The clearest way to see the difference is side by side. Claude's memory is convenient. aiperson is convenient and yours.
| Claude memory | aiperson | |
|---|---|---|
| Where it lives | Anthropic's servers | A file in your git repo + local SQLite |
| Who holds the keys | Anthropic | You (Ed25519, generated locally) |
| Works in other AIs | Claude only | Every AI at once — cloud + local |
| Moving it out | One-time, lossy export (Mar 2026) | Already there; nothing to migrate |
| Captures the AI's character | Facts about you, mostly | Voice, conventions, refusals — the persona |
| Works with local models | No | Yes — Llama, Mistral, Ollama, LM Studio |
| Export | Vendor-shaped, when offered | Any time, open format, self-hostable |
How aiperson holds your memory
- Two layers, both yours. A persona — voice, values, conventions, refusals, signed with your own Ed25519 key and kept in a git repo you own — and a corpus, a local SQLite database with hybrid search.
- Present everywhere at once. Projected into 13 editors over MCP and 21 cloud chat surfaces through a browser extension — Claude included.
- It learns as you work. Durable signal is captured into the persona as you go, so the character compounds instead of resetting.
- On your terms. A daemon reconciles roughly every five minutes; they host nothing by default; you export any time.
Frequently asked questions
What is Claude's memory?
Claude's memory is Anthropic's built-in feature that lets Claude remember details about you and your projects across conversations — your preferences, your work, the way you like answers. It is genuinely useful and improving fast. The limit is structural: it lives on Anthropic's servers, in Anthropic's format, and works only inside Claude.
Can I move my Claude memory to ChatGPT or another AI?
Since March 2026 the major vendors let you import a memory when you move in, so you can copy your Claude memory out once. But it is a one-time, lossy copy — not a living memory that stays in sync. Switch back, or run a third AI, and you are maintaining separate memories again. aiperson is different: the persona is a file you hold that every AI reads at once, so there is nothing to migrate.
Is aiperson a replacement for Claude's memory?
No — it sits above it. Keep using Claude and its memory; aiperson adds a portable persona you own that travels to ChatGPT, Gemini, local models and 30+ surfaces as well. You are not choosing one vendor's walls — you are adding a layer that none of them can take away.
Does aiperson read my Claude conversations?
Only what you opt into. aiperson captures turns through the editor MCP integration and an opt-in browser extension, and writes durable signal — a value, a convention, a refusal — into a persona you own. By default nothing is hosted: your persona is in your git repo and your corpus is a local SQLite file on your machine.
What is the un-fakeable proof that the memory is really yours?
Cultivate an AI under Claude — its voice, your conventions, the relational history — then move it intact to a locally-hosted Llama, persona and all. If the character survives the move to a model Anthropic does not run, the memory was genuinely yours. That move is impossible with vendor memory; it is the whole point of aiperson.
Keep Claude. Own the memory.
aiperson is a memory you actually hold — signed, local-first, present across Claude, ChatGPT, Gemini, local models and 30+ surfaces. Cultivate one AI, keep it yours.