v0.1.4

• Change the default cli usage
• Add cli set for easy to use govector

v0.1.3 new features update

Added

• Protobuf Storage Engine: Replaced JSON serialization with Protocol Buffers for high-performance vector and point persistence.
• Collection Auto-Discovery: The API server now automatically scans and registers all collections found in the BoltDB storage on startup.
• Data Versioning: Introduced 64-bit nanosecond-precision versioning for all points, enabling future consistency checks.
• SQ8 Vector Quantization: Integrated Scalar Quantization (8-bit) to significantly reduce disk footprint for large-scale datasets.
• Dynamic Collection Metadata: Support for persisting and loading collection-specific parameters (Dim, Metric, HNSW config).
• Safe Concurrent Access: Full mutex protection for the API server's collection registry, eliminating data races.
• Advanced Payload Filtering: Added support for Range, Prefix, Contains, and Regex filtering in both search and deletion.
• Consistent Deletion: Refactored deletion logic to ensure storage and memory index are always in sync (Storage-First-then-Memory).
• Large-Scale Benchmark Suite: New benchmark tool supporting 10K, 100K, 1M, and 10M scale testing with memory usage reporting.

Optimized

• HNSW Recall: Improved heuristic for post-filtering recall by over-fetching candidates during filtered searches.
• Test Coverage: Increased core package coverage to 92.7% and API package coverage to 92.8%.
• Memory Management: Batch-processing in benchmarks and better GC triggering for large datasets.

Fixed

• Fixed a data race in api/server.go when starting/stopping the HTTP server.
• Fixed a potential integer overflow in the SQ8 quantization algorithm.
• Fixed a bug where collections created via API were lost after server restart.

First release

Small , Fast and Free! Run it now.