The first beta of EclipseStore 4.0.0 is now available, marking a major evolution of the Java-nativ persistence engine. With this version, EclipseStore has integrated the high-performance vector similarity search engine JVector and therefore becomes an embedded Java vector database engine that runs within the same JVM process as your GenAI app. With EclipseStore 4, Java developers can now build powerful GenAI apps with pure Java, without the need to use external vector databases. This makes the entire development process of GenAI apps very convenient and significantly faster, simplifying testing and deployment.
If you later want to scale your app horizontally for larger GenAI workloads, you can run your EclipseStore app as a distributed application using Eclipse Data Grid at any time. Eclipse Data Grid runs on Kubernetes and is available On-Prem and as a fully managed PaaS service.
Both, EclipseStore and Eclipse Data Grid, are Eclipse open-source projects. The code is released under the Eclipse Public License (EPL) 2.0 that allows you to use it for commercial purposes.
Enterprises can get enterprise support for their mission-critical GenAI applications from Cyrock.AI.
Native Vector Similarity Search
JVector is a high-performance, specialized Java library built for vector similarity search. By utilizing the HNSW (Hierarchical Navigable Small World) algorithm, enabling fast approximate k-nearest-neighbor (k-NN) searches and delivers high-recall results with sub-linear search times. What sets JVector apart is its optimization for the modern JVM; it leverages the Project Panama SIMD API for lightning-fast distance calculations. Furthermore, its disk-aware design allows it to manage datasets that far exceed available RAM – aligning perfectly with the EclipseStore philosophy of efficient, large-scale data handling.
This makes EclipseStore 4.0.0 particularly well suited for modern workloads such as:
- Semantic search
- AI-driven applications
- Recommendation systems
- Embedding-based retrieval
Jvector persistence management
The GigaMap is the powerful convencience layer behind EclipseStore’s high-performance data management. With the release of version 4, we have integrated JVector directly into the GigaMap indexing architecture, creating a unified experience for Java developers.
The GigaMap now supports a specialized vector index. This means when you store your Java object graph, you can index associated vector embeddings seamlessly, benefiting from a unique, developer-centric feature set:
- Java-Native Persistence: Your vectors and business objects coexist within the same native Java storage. This eliminates the latency and serialization overhead typically found in external database calls.
- Lazy Entity Access: Search results map directly to your Java entities. Thanks to built-in lazy loading, you can access objects immediately from the result set without performing manual lookups or managing complex ID mappings.
- On-Disk Indexing: By utilizing memory-mapped files, the system ensures query performance remains high even when your index size outgrows physical RAM.
- PQ Compression: Integrated Product Quantization (PQ) allows you to reduce the memory footprint of your embeddings by up to 90%, maintaining high accuracy while significantly lowering hardware requirements.
Lucene Index Enhancements
Version 4.0.0 also introduces improvements to the Lucene integration. A new in-memory graph directory with specialized type handlers enables storing the index directly inside the persistent object graph, further strengthening the tight integration within EclipseStore. The documentation has been comprehensively updated to reflect these enhancements.
Future-Proof Native Memory Access
To ensure long-term compatibility with evolving Java versions, EclipseStore 4.0.0 introduces a high-performance native memory access layer. This abstraction safeguards performance and stability as low-level JDK memory APIs continue to evolve.
Stability Improvements and Fixes
The beta release also includes important fixes and refinements:
- Improvements for lazy reference handling to ensure storage consistency
- Various GigaMap enhancements
- Type handling updates aligned with recent JDK changes
With version 4.0.0, EclipseStore expands beyond Java object graph persistence and takes a decisive step into vector-aware data management – fully integrated, persistent, and Java-native.
Greetings from the team
If you like EclipseStore and Eclipse Data Grid and appreciate our work, please star us on GitHub. This helps us a lot and motivates us to continue our Java community work.
Resources:
Documentation:
https://docs.eclipsestore.io/manual/gigamap/indexing/jvector/index.html
Source Code:
EclipseStore:
https://github.com/eclipse-store/store
JVector Integration:
https://github.com/eclipse-store/store/tree/main/gigamap/jvector
