Skip to main content

RAG: Give Your AI Agent a Memory

Ever wished your AI agent could remember your company's documentation, search your database, or know about events that happened after its training? That's exactly what RAG (Retrieval-Augmented Generation) does.

The Problem​

Without RAG, your AI agent is like a smart person with amnesia:

// ❌ Without RAG - agent only knows training data
const agent = new Agent({
  name: "Support Bot",
  instructions: "Help customers with our product",
});

// User: "What's our return policy?"
// Agent: "I don't have access to your specific return policy..."

The Solution​

With RAG, your agent can search and use your actual data:

// ✅ With RAG - agent searches your knowledge base
const agent = new Agent({
  name: "Support Bot",
  instructions: "Help customers with our product",
  retriever: myKnowledgeBase, // 🔥 This is the magic
});

// User: "What's our return policy?"
// Agent: "According to our policy document, you have 30 days..."
// Sources: [{ title: "Return Policy", url: "docs/returns.md" }]

What You'll Build​

By the end of this guide, you'll have an AI agent that can:

  • Search through your documents instantly
  • Pull relevant info from databases
  • Give accurate answers with sources
  • Track where information came from
  • Stay up-to-date with your latest data

How It Works (2 Ways)​

VoltAgent gives you two ways to add RAG to your agents:

const agent = new Agent({
  name: "Smart Assistant",
  retriever: mySearchEngine, // Searches before every response
});

When to use: Support bots, documentation assistants, Q&A systems

Option 2: Search When Needed​

const agent = new Agent({
  name: "Smart Assistant",
  tools: [mySearchEngine.tool], // Agent decides when to search
});

When to use: General assistants, complex workflows, multi-step tasks

Works With Any Database​

The best part? VoltAgent doesn't lock you into any specific database. All retrievers extend the same BaseRetriever class, so switching is easy:

import { BaseRetriever } from "@voltagent/core";

// Start with local files
class FileRetriever extends BaseRetriever {
  async retrieve(input, options) {
    const query = typeof input === "string" ? input : input[input.length - 1].content;
    // Search local files
    const results = await this.searchFiles(query);
    return results.join("\n\n");
  }
}

// Later switch to PostgreSQL
class PostgreSQLRetriever extends BaseRetriever {
  async retrieve(input, options) {
    const query = typeof input === "string" ? input : input[input.length - 1].content;
    // Search PostgreSQL with pgvector
    const results = await this.searchPostgreSQL(query);
    return results.map((row) => row.content).join("\n\n");
  }
}

// Agent code stays exactly the same! 🎉
const agent = new Agent({
  retriever: new FileRetriever(), // Just swap the class
});

Supported patterns:

  • Vector DBs: Chroma, Pinecone, Weaviate, Qdrant
  • SQL: PostgreSQL, MySQL, SQLite
  • NoSQL: MongoDB, Redis
  • Search: Elasticsearch, Algolia
  • Files: PDF, Word, CSV, JSON
  • APIs: REST, GraphQL

Integration Examples​

Build Your Own Retriever​

Connect to your own database, API, or files with a custom retriever.

class MyRetriever extends BaseRetriever {
  async retrieve(input, options) {
    const query = typeof input === "string" ? input : input[input.length - 1].content;
    const results = await this.searchMyData(query);

    // Track sources using userContext
    if (options.userContext) {
      options.userContext.set(
        "sources",
        results.map((r) => r.source)
      );
    }

    return results.map((r) => r.content).join("\n\n");
  }
}

→ Build Your Own Retriever

Chroma Vector Database​

Perfect for beginners. Run locally with Docker and get started immediately.

# Start Chroma server
docker run -p 8000:8000 chromadb/chroma

# Try the example
npm create voltagent-app@latest -- --example with-chroma

→ Full Chroma Guide

Choose Your Path​

I want to try it now → Chroma Tutorial (10 mins)

I want to build custom → Build Your Own Retriever

I want to see examples → GitHub Examples

I need help choosing → Join our Discord and ask!

Table of Contents