Add Conversation Memory to a Chatbot (TypeScript)

mkdir shopping-assistant && cd shopping-assistant
npm init -y
npm install @neo4j-labs/agent-memory ai @ai-sdk/openai
npm install --save-dev typescript tsx @types/node

{
  "compilerOptions": {
    "target": "ES2022",
    "module": "ESNext",
    "moduleResolution": "bundler",
    "strict": true,
    "esModuleInterop": true,
    "skipLibCheck": true
  }
}

{
  "type": "module",
  "scripts": {
    "start": "tsx src/chat.ts"
  }
}

export OPENAI_API_KEY=sk-...
export MEMORY_API_KEY=nams_...

import {
  generateText,
  experimental_wrapLanguageModel as wrapModel,
  type LanguageModelV1Middleware,
} from "ai";
import { openai } from "@ai-sdk/openai";
import { MemoryClient } from "@neo4j-labs/agent-memory";
import { agentMemoryMiddleware } from "@neo4j-labs/agent-memory/middleware/vercel-ai";

const USER_ID = process.env.USER_ID ?? "demo-user";

async function main() {
  // Zero-config: reads MEMORY_API_KEY from env, targets NAMS.
  const memory = new MemoryClient();

  // One conversation per user. Reuse the same id across runs to see
  // the assistant recall context from prior sessions.
  const conv = await memory.shortTerm.createConversation({
    userId: USER_ID,
    metadata: { source: "shopping-assistant" },
  });

  console.log(`Conversation id: ${conv.id}\n`);

  // The middleware injects three-tier context before each model call
  // and persists both the user input and the assistant reply.
  const middleware = agentMemoryMiddleware(memory, {
    conversationId: conv.id,
    userId: USER_ID,
  }) as unknown as LanguageModelV1Middleware;

  const model = wrapModel({
    model: openai("gpt-4o-mini"),
    middleware,
  });

  // ...turn loop goes here in Step 3...
}

main().catch((err) => {
  console.error(err);
  process.exit(1);
});

const turns = [
  "Hi! I'm shopping for running shoes. I usually wear Nike, size 10.",
  "What's a good option for marathon training?",
  "What about something more affordable in the same brand?",
];

const system = `You are a helpful shopping assistant for an athletic-shoes
retailer. Use any preferences and prior conversation history you receive
from memory to personalize recommendations. Cite the user's stated brand
and size when relevant.`;

for (const userMessage of turns) {
  process.stdout.write(`\n[user] ${userMessage}\n[assistant] `);
  const { text } = await generateText({
    model,
    system,
    messages: [{ role: "user", content: userMessage }],
  });
  process.stdout.write(`${text}\n`);
}

npm start

Conversation id: 38f1c2e0-...

[user] Hi! I'm shopping for running shoes. I usually wear Nike, size 10.
[assistant] Great — Nike size 10 noted. Are you running on roads, trails,
or a mix?

[user] What's a good option for marathon training?
[assistant] For marathon training in Nike size 10, the ZoomX Vaporfly is
a great choice — light carbon plate, breathable upper. Want me to find
similar but for higher mileage?

[user] What about something more affordable in the same brand?
[assistant] Sticking with Nike size 10, the Pegasus 41 is the budget-
friendly daily trainer ...

// Distill the conversation into a long-term preference.
await memory.longTerm.addPreference(
  "brand",
  "Prefers Nike for running shoes",
  { context: `From conversation ${conv.id}` },
);

await memory.longTerm.addPreference(
  "size",
  "Wears US men's size 10 in athletic shoes",
  { context: `From conversation ${conv.id}` },
);

console.log(
  `\nPreferences persisted. Re-run with USER_ID=${USER_ID} to see them recalled.`,
);

USER_ID=demo-user npm start

[user] What size do I wear again?
[assistant] You wear US men's size 10 — that's what we noted last time.

// Inspect what's persisted
const recent = await memory.shortTerm.listConversations({
  userId: USER_ID,
  limit: 5,
});
console.log(`\nRecent conversations for ${USER_ID}:`);
for (const c of recent) {
  console.log(`  ${c.id} — ${c.messages?.length ?? 0} messages`);
}

const prefs = await memory.longTerm.searchPreferences(
  "running shoes",
  { limit: 5 },
);
console.log(`\nStored preferences:`);
for (const p of prefs) {
  console.log(`  [${p.category}] ${p.preference}`);
}