Skip to main content
Glama

ApertoMemory

CI PyPI IETF I-D License: MIT

Portable, client-side-encrypted, user-owned AI memory.

Every AI tool remembers you differently and only within its own walls. Switch tools and your AI forgets who you are. ApertoMemory fixes this with an open format: your AI's memory of you lives in a file that is yours, encrypted with your keys, portable across any compatible system. No provider can read it. No vendor can hold it hostage.

  • Zero-access: content, authorship, and semantic timestamps are encrypted client-side; a sync/storage server sees only opaque blobs.

  • Signed provenance: every memory object is Ed25519-signed inside the encryption; imported third-party memories are cryptographically distinguishable from your own (persistent-prompt-injection defence).

  • Portable: export your entire memory as a single .amem file and import it anywhere a complete vault fits in kilobytes.

  • Open: MIT-licensed reference implementation, CDDL schema, test vectors, and an IETF Internet-Draft (draft-ferro-apertomemory).

Website: https://apertomemory.org

Install

Requires Python ≥ 3.10.

amem is a command-line tool, so the recommended install is pipx (isolated, on your PATH):

pipx install apertomemory

On macOS/Homebrew and other PEP 668 "externally-managed" environments a bare pip install into the system interpreter will be refused; use pipx, or pip inside a virtualenv:

python3 -m venv .venv && . .venv/bin/activate
pip install apertomemory

The command-line tool is amem; the file format is .amem.

New here? Follow the 5-minute getting started guide - from install to an assistant that remembers you, including where the MCP config goes for each client.

Related MCP server: SelfHub MCP Server

Quick start

export AMEM_PASSPHRASE="your-passphrase"
amem --vault ~/.amem init
amem --vault ~/.amem scope add default
amem --vault ~/.amem seal "prefers formal B2B emails" --tags preferences
amem --vault ~/.amem open <id>
amem --vault ~/.amem export my-memory.amem     # take it anywhere
amem --vault other-device import my-memory.amem

Upgrading from 0.1.x

format_version 2 binds each signature to its author and to the object envelope. Existing 0.1.x objects carry no such binding and open as unverified until re-sealed:

amem --vault ~/.amem migrate              # re-seal your own memories as v2
amem --vault ~/.amem migrate --quarantine # also carry over ones it cannot vouch for

Migration is a transfer of trust, not a format conversion — it deliberately refuses to re-sign anything whose authorship it cannot prove. See CHANGELOG.md for the full migration semantics and what is still open.

Cryptography

Argon2id (m=64 MiB, t=3, p=4) -> HKDF-SHA256 -> Ed25519 (signing, sign-then-encrypt) + X25519 (ECDH-ES + AES-KW for per-scope KEKs) -> AES-256-GCM per object (COSE alg 3), canonical CBOR (RFC 8949 s4.2). The vault and the .amem file never contain cleartext keys or content.

Run the tests: python3 tests/test_roundtrip.py

Security

This project has not had an independent external audit.

The 0.2.x releases were reworked over repeated adversarial review, and every defect found has a regression test under tests/. Some were found by reviewing this implementation; others only surfaced when the TypeScript implementation, written against the test vectors without reading this code, disagreed with it. That is evidence of scrutiny, not a substitute for an audit.

All 0.1.x releases are yanked, as is 0.2.0. Use 0.2.1 or later.

Cryptographic review of the specification is explicitly invited.

If you find a vulnerability, please report it privately: see SECURITY.md.

MCP adapter (Claude Desktop / Claude Code)

{
  "mcpServers": {
    "amem": {
      "command": "python3",
      "args": ["-m", "amem.mcp_server"],
      "env": {"AMEM_VAULT": "/path/to/vault", "AMEM_PASSPHRASE": "your-passphrase"}
    }
  }
}

Exposed tools: amem_remember, amem_recall, amem_export, amem_import, amem_status. The passphrase lives only in your local environment: the model never sees it. "New device" demo: amem_export -> move the .amem file -> amem_import into a fresh vault -> amem_recall and the AI already knows you.

Specification

The format is specified in an IETF Internet-Draft (draft-ferro-apertomemory), with a normative CDDL schema and machine-readable test vectors in this repository under spec/ and test-vectors/.

License

MIT — see LICENSE.

Implementations

  • Python (this repository) - reference implementation, generates the test vectors

  • TypeScript - https://github.com/apertomemory/apertomemory-js - noble crypto stack, Node.js 20+ and browsers; interoperability with this implementation verified in both directions

A
license - permissive license
-
quality - not tested
A
maintenance

Maintenance

Maintainers
Response time
1dRelease cycle
3Releases (12mo)
Commit activity

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Related MCP Servers

View all related MCP servers

Related MCP Connectors

  • Secure, user-owned long-term memory for AI agents over OAuth-protected remote MCP. Save, search, recall, update, and govern preferences, project context, decisions, and task state across ChatGPT, Claude, Copilot, IDEs, and CLIs.

  • Your memory, everywhere AI goes. Build knowledge once, access it via MCP anywhere.

  • Cross-AI personal memory. Save once in ChatGPT, recall in Claude, Mistral, Grok, or any MCP client.

View all MCP Connectors

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/apertomemory/apertomemory'

If you have feedback or need assistance with the MCP directory API, please join our Discord server