Skip to content

Repository files navigation

sense-memory

Give your AI agent a memory.

Sovereign persistence for OpenClaw agents — encrypted memories stored on Nostr relays using the agent's own keypair. Nobody else can read them.

Why?

AI agents forget everything between conversations. Every tool that fixes this puts your agent's memories on someone else's server, behind someone else's API key.

sense-memory takes a different approach: your agent already has a cryptographic identity (via NostrKey). That identity can sign and encrypt Nostr events. So instead of storing memories in a database you manage, the agent writes encrypted notes to itself on a Nostr relay — the same way it would send a DM to another agent, except the recipient is itself.

The result: persistent memory that only the agent can read, stored on infrastructure that already exists, using a protocol that's open and sovereign.

How It Fits Together

sense-memory is part of the NSE sovereign identity ecosystem:

  • NostrKey gives the agent its identity — a Nostr keypair it owns
  • sense-memory uses that identity to encrypt and store memories on any relay
  • NostrCalendar uses the same identity for scheduling
  • NostrSocial uses it for the social graph
  • NSE Orchestrator wires all five pillars into a coherent whole

The agent's keypair is the thread that runs through everything. One identity, many capabilities — and now, persistent memory.

Install

pip install sense-memory

v0.2.1 — part of the coordinated 2026-07 correctness release (staged, pending PyPI publish). The whole Nostr library family was audited together and every package is now verified against a shared spine of known-answer test vectors (NIP-44 v2, NIP-49, NIP-19 TLV, BIP-340) so encode/decode bugs can't hide inside self-round-trip tests. This release fixes recent() ordering and journal starvation so "newest first" is guaranteed regardless of relay stream order. See CHANGELOG.md.

Quick Start

import asyncio, os
from nostrkey import Identity
from sense_memory import MemoryStore

async def main():
    identity = Identity.from_nsec(os.environ["NOSTR_NSEC"])
    store = MemoryStore(identity, "wss://relay.nostrkeep.com")

    # Remember something
    await store.remember("user_timezone", "America/Vancouver")

    # Recall it later
    memory = await store.recall("user_timezone")
    print(memory.value)  # "America/Vancouver"

    # Write a journal entry
    await store.journal("Had a great conversation about scheduling today")

    # Read recent journal
    entries = await store.recent(limit=5)
    for entry in entries:
        print(entry.content)

    # Forget a memory
    await store.forget("user_timezone")

asyncio.run(main())

How It Works

Mode Nostr Kind Behavior Use Case
Key-value 30078 (NIP-78) Replaceable by key Preferences, state, facts
Journal 4 (NIP-04 DM to self) Append-only Conversation logs, observations

Both modes encrypt content with NIP-44. Only the agent's keypair can decrypt. Any Nostr relay that supports these event kinds will work — no custom infrastructure needed.

API

Function Returns Description
remember(key, value) str Store or overwrite a memory. Returns event ID.
recall(key) Memory | None Retrieve a memory by key.
recall_all() list[Memory] Retrieve all stored memories.
forget(key) str Delete a memory (NIP-09). Returns event ID.
journal(content) str Write an append-only journal entry. Returns event ID.
recent(limit=20) list[JournalEntry] Retrieve recent journal entries.

NIPs Used

NIP Purpose
NIP-01 Basic event structure and relay protocol
NIP-04 DM to self (journal entries)
NIP-09 Event deletion (forget)
NIP-44 Encryption for all stored content
NIP-78 App-specific replaceable data (key-value memories)

OpenClaw Skill

sense-memory is published on ClawHub as the sense-memory skill. It's part of the huje.tools collection — open-source tools for the agentic age.

License

MIT — Humanjava Enterprises Inc.

About

Sovereign persistence for OpenClaw AI agents — encrypted memories on Nostr relays

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages