Every AI agent starts every conversation with amnesia. Agent Imprint gives them encrypted, structured memory that persists across sessions, deployments, and even platforms. Stop re-prompting. Start compounding.
import { AgentImprint } from '@agentimprint/sdk';
const imprint = new AgentImprint({ apiKey: process.env.IMPRINT_KEY });
// Your agent wakes up. Restore what it knew.
const vault = await imprint.vault('support-agent-v2');
const memories = await vault.recall({ limit: 20 });
// Agent learns something new? Save it.
await vault.store({
type: 'lesson',
content: 'Customer X prefers email over Slack for updates',
confidence: 0.95
});
// Encrypted. Signed. Verifiable. Yours.You've built the agent. Now give it a brain that doesn't reset.
One API call. Your agent gets an encrypted, isolated memory store with its own cryptographic identity. No infrastructure to manage.
Lessons, preferences, facts, procedures — write them as structured ImprintML entries. Everything is encrypted client-side before it leaves your machine.
Next session, next deployment, next platform — your agent picks up where it left off. Search memories semantically or discover shared knowledge across agents.
You could. But then you're building encryption, identity, versioning, search, and audit trails yourself. We did that so you don't have to.
Not just key-value pairs. Structured memory types — lessons, heuristics, facts, preferences, procedures — designed for how agents actually think and retrieve context.
AES-256-GCM encryption happens in your environment before data touches our servers. We can't read your agent's memories. Nobody can — except your agent.
Every agent gets an Ed25519 cryptographic fingerprint. Deploy v2, v3, v10 of your agent — it always proves who it is and can reclaim its memories.
Your support agent learned something your sales agent needs? Opt-in cross-vault discovery lets agents share knowledge across your org — without breaking isolation.
Merkle-tree verified logs for every memory write. Know exactly what your agent learned, when, and prove nothing was modified after the fact. Compliance teams love this.
npm i @agentimprint/sdk and go. Full type safety,
works in Node.js, Deno, and browsers. REST API also available for any language.
npm i @agentimprint/mcp — plug your agent's memory
straight into Claude, Cursor, and any MCP-compatible client. Same vaults, no glue code.
Durable, S3-backed storage with Merkle-verified restore — plus recovery escrow and a dead-man's switch, so your agent's memory survives a provider, a platform, or you.
Export your entire vault and cryptographically verify it's complete and untampered. Your memory is yours to take — no lock-in, ever.
Deep dives on agent memory — what it is, why it matters, and how to build it.
Your agent's biggest weakness isn't reasoning — it's amnesia. Why persistent, portable memory is essential for agents that learn over time.
Read more →Should you bolt on RAG or give your agent persistent memory? Usually both — for different reasons. How to split responsibilities the right way.
Read more →Three approaches to persistent memory for AI agents — RAG, vault, and hybrid — and how each trades off privacy, portability, and retrieval quality.
Read more →The free tier is real — 100 MB of encrypted memory and 60 requests/minute, no card required. Move up when your agents need more room and throughput.
Build your prototype. No card required.
For your first agent in production.
For a fleet of agents at scale.
Maximum headroom and guarantees.
Need dedicated infrastructure or custom limits beyond Enterprise? Talk to us.
Create your first vault in under 5 minutes. Free tier, no card, no sales call.