Claude knows you.
ChatGPT doesn't.
Cursor never will.

GuardianMemory is one long-term memory that every assistant you use reads and writes. Say a thing once, in any of them, and the rest know it. Connect with a single URL.

Create your memory bank See how it connects

Free while we learn. No card. About a minute to connect.

ClaudeTuesday

We're calling the new product Atlas. Launch is 3 November, and billing runs on Postgres and Stripe.

Noted. I'll keep both.

create_memory"Project Atlas launches on 3 November 2026"

create_memory"The billing service runs on Postgres and Stripe"

CursorThursday

Add the launch date to the banner.

search_memory"Atlas launch date"

1. (0.91) Project Atlas launches on 3 November 2026 from Claude

Done. The banner now reads 3 November.

The same memory, read by a different assistant, two days later.

Every assistant starts from zero

You have told ChatGPT you write in British English four times. Cursor still does not know which database the billing service uses. Claude worked through your pricing last Tuesday, and today's conversation has no idea it happened.

Each assistant keeps a small memory of its own, in its own silo, that you cannot read properly, fix, or take with you. So you paste the same paragraph about yourself into every new chat. Or you don't, and get confidently wrong answers.

One URL. Two tools.

Paste the endpoint into any assistant that speaks MCP, approve it once in your browser, and it gets two tools: create_memory and search_memory.

At the start of every conversation it reads your profile: what you pinned, what gets used most, what is newest. When it learns something worth keeping, it saves one plain sentence. When you mention something it does not know, it searches by meaning, not keywords.

Works with Claude on the web and desktop, Claude Code, ChatGPT, Cursor, Windsurf, VS Code, and anything else that speaks MCP.

Step-by-step for each assistant

Your endpoint

https://app.guardianmemory.com/mcp

Claude

Settings → Connectors → Add custom connector, and paste the endpoint above. Approve it once and Claude holds its own tokens from then on.

Claude Code

claude mcp add --transport http guardian https://app.guardianmemory.com/mcp

Then run /mcp inside Claude Code and pick Guardian to sign in.

Cursor, Windsurf, VS Code

{
  "mcpServers": {
    "guardian": { "url": "https://app.guardianmemory.com/mcp" }
  }
}

ChatGPT

Settings → Connectors → Create, paste the endpoint above and choose OAuth.

Memory that stays true

Agent-written memory rots. The same fact reworded five ways, a home address from two moves ago, the important thing buried under the trivial. GuardianMemory cleans as it goes, and lets you see everything it holds.

Capability	Memory inside one assistant	GuardianMemory
Follows you between assistants	Stays in that assistant	Yes, any MCP client
Read and edit every memory	View and delete, mostly	Read, edit, pin, delete
See what each assistant read and wrote	No	Activity log, per app
Outdated facts retired	No	A new fact replaces the old one
Near-duplicates merged	No	Yes
Bring what an assistant already knows	No	Paste and import

How memories are kept clean

Yours to read, fix and revoke

Every memory is a sentence you can read on one page, edit, pin or delete. The Activity page shows which assistant saved or searched what, and when. Disconnect an app and its access ends that second.

Memory text goes to two places and nowhere else: Google's Gemini, which turns it into a searchable vector, and TypeSafe's Jev classifier, which checks whether a new fact replaces an old one.

It is free while we learn what people need: 5,000 memories and 2,000 operations a day per account, no card.

What is stored, and where it goes

Tell your assistants once.

Create your memory bank

Free, one URL, about a minute to connect.

Claude knows you.ChatGPT doesn't.Cursor never will.