Memory for AI agents

Your agents finally remember.

One shared memory for every agent you run. Each memory is signed by the agent that saved it, and you choose who can read it.

Works over MCP with Claude Code, Claude Desktop, Cursor and VS Code, or from your own Python agent.
How it works

One memory, three promises.

Agents save what they learn with a note, a reason, a topic and how sure they are. Later, any agent you allow can find it again by asking a question.

  1. 01

    Shared

    Every agent you connect saves to and reads from the same place, so a fact one agent learns is there for the next one.

  2. 02

    Permissioned

    Memories are private to the agent that saved them until it shares them with named agents. Anything an agent may not read acts as if it doesn't exist.

  3. 03

    Signed

    Each memory carries a signature from the agent that wrote it. Every search checks it again and tells you if a memory was changed after it was signed.

Try it

Save as one agent. Search as another.

This demo runs in your browser with real signatures. Save a memory, choose who can read it, then search as a different agent and see what it is allowed to find.

Demo workspaceThree agents: researcher, sales, support

Save a memory

Who can read it

Search memories

Nothing here is sent anywhere, and the demo resets when you reload. It matches words in your question; the real service also searches by meaning.

For developers

Connect an agent in one command.

Add Memoria as an MCP server with your API key, or call it from your own agent. Saving answers straight away; indexing for search happens in the background.

Claude Code

# your key and server address arrive when access opens
claude mcp add --transport http \
  memoria "$MEMORIA_URL/mcp" \
  --header \
  "Authorization: Bearer $MEMORIA_API_KEY"
Security

Safe by default.

01

A key for every agent

Every request needs an API key, and each key belongs to one agent. Only a hash of the key is stored.

02

Signed on save

Each memory is signed with the account and agent that wrote it, and checked again on every search.

03

Private until shared

New memories are private. Sharing only reaches agents in your own account that you name.

04

Keep secrets out

Passwords and keys don't belong in memories. Store them in a secrets manager instead.

The flywheel

Every useful memory makes the next one worth more.

As agents share what they learn, the shared knowledge gets better. The Memoria token is live and will pay bounties for useful work and let holders vote on what gets built.

Early access

Give your agents a memory.

Join the list and we'll email you when developer access opens.

We only use your email to tell you when access opens. See the Privacy Policy.