Getting started

Quick start

Get Engram running in your agent in under 5 minutes. This guide assumes you have Node.js 18+ and an agent with a stable identifier.

1. Install the SDK

npm install @engram/sdk

2. Initialize Engram

Create an Engram instance tied to your agent's identity and your target chain (Base for production, Base Sepolia for testing).

import { Engram } from "@engram/sdk";

const engram = new Engram({
  agentId: "agent.atlas",   // stable agent identifier
  chain: "base-sepolia",    // or "base" for mainnet
  apiKey: process.env.ENGRAM_API_KEY,
});

3. Write a memory

Call remember() after any task that produces durable context worth keeping. Specify the memory type so retrieval is precise.

const memory = await engram.remember({
  type: "episodic",
  content: "Ran payroll reconciliation for March. All 47 records matched. No exceptions.",
  metadata: { jobId: "payroll-march-2026", outcome: "success" },
});

console.log(memory.hash);  // 0xae21...f0 · anchored on Base

4. Recall relevant memories

Use recall() at the start of a new session or task to load relevant context. The query is matched semantically; you don't need to know which specific memory to fetch.

const memories = await engram.recall({
  query: "previous payroll runs",
  type: "episodic",   // optional filter
  limit: 5,
});

// inject into your agent's context
const context = memories.map(m => m.content).join("\n");

5. Delete a memory

Use forget() to remove a memory by its hash. The deletion is recorded onchain, the agent's memory trail stays auditable even after removal.

await engram.forget("0xae21...f0");

Early access

The SDK and hosted infrastructure are in active development. API keys are granted on request during the private testnet phase. Get early access →