Overview

How it works

Databaset sits between your app and your AI model. We handle memory storage and retrieval so you can focus on the product.

The pipeline

1. You send text

Your app sends us raw text: messages, preferences, facts, anything worth remembering.

await memory.store({
  userId: 'user_123',
  text: 'User prefers dark mode and concise responses'
})

2. Auto chunking

We split large text into optimal chunks. You do not need to worry about token limits or chunk sizes.

3. Embedding

Each chunk becomes a vector embedding. Similar meanings end up close together in vector space.

4. Storage

We store embeddings and original text with userId, appId, timestamps, and any metadata you pass.

5. Smart retrieval

When you recall, we find semantically relevant memories, not just keyword matches. Recent memories get priority. We also handle contradictions automatically.

6. Context ready

You get back a clean context string, ready to drop into your AI prompt.

const context = await memory.recall({
  userId: 'user_123',
  query: 'user interface preferences'
})
// "User prefers dark mode. User wants concise responses."

Next steps

• Key Concepts: memories, users, apps
• Retrieval deep dive: how semantic search works