End-to-end, every payload
NIP-44 v2 encryption keyed to your nsec. We never see your plaintext, your embeddings, or your queries. Server-side backups stay encrypted at rest.
Agentic memory · routed to Nostr
The memory layer that outlives the model.
Nostrmind is a paid API for portable, end-to-end encrypted AI memory. Managed keys by default. Bring your own key when you want it. Routed across the open Nostr relay network so your agent's memory follows it everywhere.
Built to plug into the agents you already ship
Cursor
Claude Code
GPT-5
LangGraph
Mastra
Vercel AI SDK
Why Nostrmind
Encryption
Portability
For agents
Vault
Backups
Open
A memory layer that survives the model, the vendor, and the platform.
Every memory is redacted, embedded, encrypted to your key, signed, and fanned out to multiple Nostr relays. The plaintext never leaves your control. The vector index is yours to rebuild on any device.
Survives any vendor change
Your memory lives on signed Nostr events. Move providers, change agents, stay sovereign. We can't lock you in.
One API key. Every agent.
Idempotent writes, fast vector queries, incremental sync, encrypted backup. Ready for Cursor, Claude Code, GPT, and your own loop.
Managed key custody
AES-256-GCM envelope, exportable nsec.
Encrypted snapshots
Pro+ adds an encrypted server mirror with one-click restore.
Read with any Nostr client
Memories are kind: 1986 events. No proprietary store.
nostr-tools subscribe \
--kinds 1986 \
--authors $YOUR_NPUB
Live API
Three calls. Your agent has memory.
No SDK to install. Standard HTTP, idempotent writes, and vector search out of the box. Drop it into Cursor, Claude Code, or any backend.
~/agent/memory.ts
// Write a memory once
await fetch("https://nostrmind.com/v1/memory/write", {
method: "POST",
headers: { authorization: `Bearer ${KEY}` },
body: JSON.stringify({
text: "User prefers dark mode",
tags: ["ui-pref"],
idempotencyKey: "mem-001",
}),
});
// Recall it from anywhere
const r = await fetch(".../v1/memory/query", {
method: "POST",
headers: { authorization: `Bearer ${KEY}` },
body: JSON.stringify({ query: "theme preference", topK: 3 }),
});
How it works
From raw text to encrypted memory in one call.
Your text passes through four steps before it ever touches a relay. Everything but the last step happens on your behalf — encrypted to a key only you can use.
1
Redact
PII pulled out and replaced with placeholders. Mappings inlined in the encrypted payload.
2
Embed & quantize
Vectorize with your model of choice, then int8-compress for compact relay payloads.
3
Encrypt & sign
NIP-44 v2 encrypt the payload, sign as a kind-1986 Nostr event with your key.
4
Fan out
Publish in parallel to multiple relays, with retry, backoff, and per-relay health metrics.
Pricing
Pick a plan that fits your agent.
Real prices. No seats. Upgrade for encrypted server backups and faster restores.
Pro API
$10/month
Everything you need to give one agent durable memory.
- Unlimited memories on shared relays
- Managed key (BYOK optional)
- Encrypted client export
- Priority Nostr fanout
Pro+ Backup
$20/month
Encrypted server backup mirror plus faster restores.
- Everything in Pro
- Encrypted server backup snapshots
- One-click restore
- Relay health metrics
Team
Custom
For products putting memory in production at scale.
- Everything in Pro+
- Audit logs & SSO
- Dedicated relays
- Annual contract
FAQ
Common questions
Can you read my memories?
No. Every payload is NIP-44 encrypted with a key derived from your account's nsec. We see only ciphertext, signed event ids, and relay-routing metadata.
What if Nostrmind disappears tomorrow?
Your memories are signed Nostr events on relays you can reach with any client. Export your nsec from the dashboard at any time. The CLI in this repo can rebuild your vector index without us.
Do you support Bring Your Own Key?
Yes. Import your own nsec at signup or rotate to a key you control later. Managed keys are the default for ease of use.
Why Nostr instead of S3 or your own DB?
Nostr gives us a censorship-resistant, multi-vendor relay network. Your memory becomes truly portable instead of locked to one provider. We also pay relay operators a share of revenue.
How fast is vector query?
Sub-100ms warm for ~50k 512-dim vectors on a single core. We use brute-force cosine search; no ANN index, no surprise misses.
Do you support team accounts?
Yes — the Team plan adds shared spaces, audit logs, and dedicated relays. Talk to us.