Personal memory for AI

Tell ChatGPT once. Claude already knows.

Everything you have told ChatGPT, Claude, or Notion, kept in one memory you own and read by every app you connect, your code editor included. Open it. Correct it. Take it with you.

WORKS WITH
Papez
Claude
ChatGPT
Notion
Cursor
Windsurf
VS Code
Zed
How to connect →
WHAT IT DOES

One memory, everywhere.

The memory ChatGPT keeps stays in ChatGPT. Claude's stays in Claude. Neither reads the other, and neither shows you why it remembers. Papez sits underneath all of them: say something once and every connected AI can recall it, tell you why, and let it go when it stops mattering.

Follows you.

Claude, ChatGPT, and Notion today; Claude Code, Cursor, and every other MCP app with the same sign-in.

Explains itself.

Ask why it remembers. You get the actual chain behind a memory, how it connects to everything else, not a guess.

Forgets on purpose.

A memory is only let go when its score has decayed, nothing links to it, and you have not pinned it. Anything connected stays.

score = relevance × connectivity × reactivation
OWNERSHIP

Yours. Check for yourself.

A memory you can open up. Inspect the graph, correct anything, export all of it, or delete it for good, and read the engine that does it.

InspectSee every memory and the causal chain behind it.
Amend"Actually, I moved." The correction supersedes the old memory, and every connected app sees the new one.
ExportTake your whole graph with you, every memory and every link, as one JSON file.
EncryptedAES-256-GCM at rest, one key per user. A stolen copy of the database is unreadable. The service decrypts to search, so we do not call this zero knowledge.
DeleteDelete any memory for good. Erase the whole account with one email; a button is on the roadmap.
Not trainedNo model is trained on your memories, ours or anyone else's. The processors that see text are listed, with what each one does.
Not yet: SOC 2, an erase button in the app, a managed key service. Each is on the roadmap; the security page changes when they ship.
PROOF

We measure. We publish. Including our misses.

85.55 on LoCoMo, the standard test of whether an AI remembers a long conversation. The mean of ten runs under a frozen protocol: same judge, same answerer, same 1,540 questions for every system we compare against. We publish what we lost as well as what we gained.

85.55
LoCoMo · ± 0.37 · July 2026
+10.4
vs Zep 75.14, same protocol
10
runs · frozen judge and answerer
Figures measured under other judges or answerers are not comparable. Disclosures
Gained170 questions the baseline cannot answer, on every run.
Lost72 the baseline got right. Net +98, published with the number, not behind it.
PRICING

300 memories a month, free. Reading is never gated.

Start free with the full engine. Founding Members pay $8 a month, billed monthly, locked for as long as they stay. After the first 500, Pro is $10 a month, or $8 billed annually.

Free$0/mo
The full engine

300 new memories a month with scoring, the causal graph, and forgetting. Reading is never gated.

FIRST 500 · ENDS DEC 31, 2026
Founding Member$8/mo
Cap lifted, price locked

Everything in Pro at $8/mo instead of $10, for as long as you stay. First 500 members or Dec 31, 2026, whichever comes first.

Join the waitlist
QUESTIONS

Straight answers.

What is Papez?

One memory you own, shared by every AI you use. Everything you tell ChatGPT, Claude, or Notion is kept in a single causal graph, and every app you connect reads the same one.

How is this different from ChatGPT's or Claude's built-in memory?

Built-in memory lives inside one app, and neither reads the other. Papez sits outside both. Say something once, every connected app recalls it, and you can see the chain that explains why it remembers.

Which apps does it work with?

ChatGPT, Claude, and Notion today. Claude Code, Cursor, VS Code, Windsurf, Zed, and any other MCP app with the same sign-in.

Is it open source?

The engine is. papez is AGPLv3 on GitHub and PyPI and runs as a stdio MCP server on your own machine. The hosted service adds sync, encryption at rest, and the connectors.

What does it cost?

Free: 300 memory stores a month, and reading is never gated. Founding Member: $8 a month, locked for as long as you stay, first 500 members, ends Dec 31, 2026. Pro: $10 a month, or $8 a month billed annually.

Can I leave?

Yes. Export the whole graph, every memory and every link, as one JSON file, and delete any memory for good. Account erasure is by email today; a button is on the roadmap.

How accurate is recall?

85.55 on LoCoMo, the mean of ten runs with a frozen gpt-4o-mini answerer and judge at temperature 0, n = 1,540. Per-run results and the questions we got wrong are published.

FOR DEVELOPERS

Papez runs on papez, our open-source engine, AGPLv3. pip install papez · stdio MCP server · pluggable storage.

Named for the Papez circuit, the loop in the brain linked to episodic memory.

Tell it once. Every app you connect remembers.

Free for 300 memories a month. Reading is never gated. Founding Member $8/mo for the first 500, through Dec 31, 2026.