Add Amazon ElastiCache / Valkey as a Vector Store

Expected Behavior

Valkey/ElastiCache should be usable as a VectorStore for RAG applications, semantic search, and semantic caching using the official Valkey Glide Java client. I am willing to contribute this implementation.

// wiring example
@Bean
public GlideClient glideClient() {
    return GlideClient.createClient(/* config */).get();
}

@Bean
public VectorStore vectorStore(GlideClient glideClient, EmbeddingModel embeddingModel) {
    return ValkeyVectorStore.builder(glideClient, embeddingModel)
        .indexName("spring-ai-index")
        .build();
}

// usage example
@Configuration
public class ValkeyConfig {

    @Bean
    public ValkeyVectorStore vectorStore(
        @Value("${spring.ai.valkey.endpoint}") String endpoint,
        EmbeddingModel embeddingModel) {

        return new ValkeyVectorStore(endpoint, embeddingModel);
    }
}

@Service
public class DocumentService {

    @Autowired
    private ValkeyVectorStore vectorStore;

    public List<Document> search(String query) {
        // Semantic search with metadata filtering
        SearchRequest request = SearchRequest.query(query)
            .withTopK(5)
            .withSimilarityThreshold(0.7)
            .withFilterExpression("source == 'docs' && year >= 2024");

        return vectorStore.similaritySearch(request);
    }
}

The implementation would use:

• FT.CREATE / FT.SEARCH commands for vector indexing and search
• HNSW and FLAT vector indexing algorithms
• TAG, NUMERIC field types for metadata filtering

Current Behavior

Spring AI supports Redis as a VectorStore using the Jedis client. While Valkey is Redis-compatible at the protocol level (RESP), there is no integration using the official Valkey Glide client.

Key differeces that justify a separate implementation:

• Different client library: Glide has a fundamentally different architecture (async Rust core with Java bidnings) compared to Jedis (pure Java)
• Native cluster support: Glide provides transparent request routing for ElastiCache cluster deployments
• AWS-maintained: Glide is the official client for ElastiCache/Valkey, actively maintained by AWS

Context

Valkey is an open-source, Redis-compatible in-memory data store. Amazon ElastiCache offers Valkey as a managed service. Both support vector similarity search.

This feature would benefit users who:

• Use Amazon ElastiCache with Valkey
• Want to use the official AWS-maintained Glide client
• Need sub-millisecond query latency for RAG applications and semantic caching

The existing spring-ai-redis-store with Jedis could be used as a workaround since Valkey is protocol-compatible, but this misses the benefits of the purpose-built Glide client.

Start by reading spring-ai-redis-store/src/main/java/org/springframework/ai/vectorstore/redis/RedisVectorStore.java and the Valkey Glide Java client documentation linked in the issue. Define the integration around FT.CREATE and FT.SEARCH, supporting HNSW and FLAT indexes plus TAG and NUMERIC metadata filters; done means Valkey/ElastiCache can provide the VectorStore behavior described for RAG and semantic search.