Build a Private LLM Knowledge Base with Redis and DeepSeek4J in 10 Minutes

Dual Function Overview

As a Cache System

Redis is best known for its caching capabilities. It can:

Fast key‑value access

Support multiple data types (strings, hashes, lists, etc.)

Provide data expiration mechanisms

Deliver high‑performance read/write operations

As a Vector Database

Redis can also serve as a vector database, supporting:

Storage of high‑dimensional vector data

Similarity search (KNN queries)

Vector range queries

Metadata filtering

Running Test

<code>docker run -p 6379:6379 --name redis-stack redis/redis-stack:latest</code>

Vector Data Storage

Redis supports two main vector storage methods:

Hash storage

<code>HSET docs:01 doc_embedding &lt;vector_bytes&gt; category sports</code>

JSON storage

<code>JSON.SET docs:01 $ '{"doc_embedding":[0.34,0.63,-0.54,-0.69,0.98,0.61], "category": "sports"}'</code>

When visualizing data with the RedisInsight client, users can see stored vectors and related information. Note that the KEYS command cannot display vector data because vectors are stored in a specific format within hashes or JSON.

Java SDK Integration

This article uses Lettuce as the Java client to interact with Redis vectors.

<code>import io.lettuce.core.RedisClient;
import io.lettuce.core.api.StatefulRedisConnection;
import io.lettuce.core.api.sync.RedisCommands;
/**
 * @author lengleng
 */
public class LettuceVectorExample {
    public static void main(String[] args) {
        // Create Redis client
        RedisClient redisClient = RedisClient.create("redis://localhost:6379");
        StatefulRedisConnection<String, String> connection = redisClient.connect();
        RedisCommands<String, String> commands = connection.sync();

        // Create index with vector field
        String createIndex = "FT.CREATE myIndex ON HASH PREFIX 1 doc: SCHEMA embedding VECTOR FLAT 6 TYPE FLOAT32 DIM 4 DISTANCE_METRIC COSINE";
        commands.sendCommand("FT.CREATE", "myIndex", "ON", "HASH", "PREFIX", "1", "doc:", "SCHEMA", "embedding", "VECTOR", "FLAT", "6", "TYPE", "FLOAT32", "DIM", "4", "DISTANCE_METRIC", "COSINE");

        // Store vector data
        float[] vector = {0.1f, 0.2f, 0.3f, 0.4f};
        String vectorStr = floatArrayToString(vector); // Convert to string representation
        commands.hset("doc:1", "embedding", vectorStr);

        // Execute vector search (KNN)
        float[] queryVector = {0.15f, 0.25f, 0.35f, 0.45f};
        String query = "(@embedding:[VECTOR_RANGE 0.5 " + floatArrayToString(queryVector) + "])=>{$KNN: 1}";
        String result = commands.sendCommand("FT.SEARCH", "myIndex", query).toString();
        System.out.println("Search result: " + result);

        // Close connection
        connection.close();
        redisClient.shutdown();
    }

    // Helper method: convert float array to space‑separated string
    private static String floatArrayToString(float[] vector) {
        StringBuilder sb = new StringBuilder();
        for (float v : vector) {
            sb.append(v).append(" ");
        }
        return sb.toString().trim();
    }
}
</code>

Conclusion

Redis's dual functionality provides unique advantages for modern application architectures. By configuring and using it properly, a single instance can simultaneously satisfy caching and vector‑search needs, saving resources and simplifying system design.