Step 5: Add Memory

Step 5: Add Memory

Right now, if you close the browser and come back, the agent has no memory of your previous conversation. Let’s fix that.

How Memory Works

The framework supports three memory types:

Type	Behavior	Use Case
none	No memory, every request is independent	Stateless tools, one-shot queries
conversational	Remembers within a thread, expires after TTL	Chat bots, support agents
persistent	Long-term storage, searchable	User preferences, learned facts

We already set memory: { type: 'conversational', ttl: '7d' } on our agent. This means:

• Messages within a conversation thread are remembered
• After 7 days of inactivity, the memory is cleared

Define a Memory Schema

The memory/ directory holds structured memory schemas. Open memory/conversations.ts:

memory/conversations.ts

import { defineMemory } from '@kernova/framework';

export default defineMemory({
  name: 'conversations',
  schema: {
    summary: { type: 'text', indexed: true },
    userId: { type: 'text', indexed: true, required: true },
    sentiment: { type: 'text' },
    topics: { type: 'json' },
    resolved: { type: 'boolean' },
  },
  ttl: '30d',
  autoSummarize: true,
});

This schema tells Kernova what to remember about each conversation:

• summary — A text summary of what was discussed
• userId — Who was talking (for per-user memory)
• sentiment — Was the user happy, frustrated, etc.
• topics — What subjects came up (as JSON array)
• resolved — Was their issue resolved?

Using Memory from the Frontend

You can also interact with memory from your React components using the useMemory hook:

'use client';
import { useMemory } from '@kernova/framework/react';

export default function PreviousChatsPage() {
  const { entries, search, isLoading } = useMemory('conversations');
  const [query, setQuery] = useState('');

  const handleSearch = () => {
    search(query, 10); // search for up to 10 results
  };

  return (
    <div>
      <h2>Previous Conversations</h2>

      <input
        value={query}
        onChange={(e) => setQuery(e.target.value)}
        placeholder="Search past conversations..."
      />
      <button onClick={handleSearch} disabled={isLoading}>Search</button>

      {entries.map((entry) => (
        <div key={entry.id}>
          <p>{entry.content}</p>
          <small>{entry.createdAt}</small>
        </div>
      ))}
    </div>
  );
}

Test Memory Persistence

1. Open http://localhost:3000/chat
2. Have a conversation: “What are the pricing plans?”
3. Note the agent’s response
4. Send another message in the same thread: “Which one do you recommend for a team of 5?”

The agent should remember you were asking about pricing and give a recommendation based on the context — it didn’t need you to repeat yourself.

What You Learned

• Memory is configured per-agent with the memory field
• conversational memory auto-tracks the conversation thread
• defineMemory creates structured schemas for stored knowledge
• useMemory hook lets the frontend read/search memory
• The TTL (7d, 30d, etc.) controls how long memories persist

---

Continue to Step 6: Deploy to Production →