0.0.3

0.0.3 (General Availability)

New Features

Index Types and Transformations

• IndexPreTransform: Modular vector transformation framework with standardized API
  • L2NormTransform for unit-length normalization
  • PCATransform for dimensionality reduction
  • RemapDimensionsTransform for feature selection
• Binary Vector Support: Comprehensive binary index implementations
  • BinaryIndex base class with efficient Hamming distance calculations
  • IndexBinaryFlat for exact binary vector search
  • IndexBinaryIVF for fast approximate search with inverted file structure
  • IndexBinaryHash for hash-based binary vector lookup
• Core Index Optimizations: Enhanced implementations of key index types
  • IndexPQ with robust vector extraction and fallback strategies
  • IndexIVFScalarQuantizer with improved training strategies
  • IndexIDMap and IndexIDMap2 for custom vector IDs with batched operations

Advanced Operations

• Factory Pattern: Create indices from string descriptions using index_factory
• Index Persistence: Direct index save/load with write_index/read_index
• Index Modification: Merge and split indices with optimized vector operations
• Vector Reconstruction: Perfect floating-point precision reconstruction
  • Single vector reconstruction (reconstruct)
  • Batch reconstruction (reconstruct_n) for efficiency
• Vector Caching: Intelligent caching across all index implementations
• Batched Processing: Optimized handling of large vector operations

Optimization Controls

• Search Parameters: Fine-grained control over nprobe, efSearch, k_factor
• Training Parameters: Configurable n_iter, min_points_per_centroid
• HNSW Parameters: efConstruction and search-time controls
• Memory Management:
  • Memory mapping for large indices
  • Usage limits and tracking with configurable thresholds
  • Automatic unloading of unused indices
  • I/O buffer size controls

Authentication & Security

• Multi-Tenant Authentication: Production-ready authentication system
  • Support for CLI auth-keys format ("key1:tenant1,key2:tenant2")
  • JSON auth-file format support
  • Complete tenant data isolation
  • Comprehensive logging of authentication events

Reliability & Error Handling

• Connection Resilience: Robust error handling and recovery
  • Configurable retry attempts with exponential backoff
  • Automatic reconnection on network failures
  • Connection health monitoring with event callbacks
  • Graceful degradation for server connectivity issues
• Enhanced Error Recovery: Improved handling of common error conditions
• Server Compatibility: Better compatibility across different server implementations

Bug Fixes

• Authentication: Fixed critical server-side authentication enforcement
• Vector Reconstruction: Resolved architectural issues affecting precision
• API Consistency: Fixed method inconsistencies in IVF-PQ implementation
• Training Behavior: Corrected scalar quantizer training to match FAISS behavior
• Environment Variables: Fixed server port configuration handling

Documentation & Testing

• Comprehensive Test Suite: Full coverage of optimized implementations
• Enhanced Examples: Extended example suite demonstrating advanced features
• API Documentation: Improved documentation for advanced usage patterns
• Authentication Tests: Complete test coverage for multi-tenant scenarios

0.0.2

0.0.2

Initial release of FAISSx, a high-performance vector database proxy using FAISS and ZeroMQ.

Added

Project Infrastructure

• Project renamed from FAISS-Proxy to FAISSx
• Directory structure reorganized (faissx, client, server, examples, data)
• Build system configured (setup.py, MANIFEST.in)
• Documentation updated
• Basic Docker deployment

Server Implementation

• ZeroMQ server application structure
• Authentication with API keys
• FAISS manager for vector operations
• Binary protocol for CRUD operations on indices
• Vector addition and search operations
• Tenant isolation for multi-application deployments
• Docker container setup
• Comprehensive server documentation with API protocol details

Client Implementation

• Client package structure
• Configuration management
• Remote API client using ZeroMQ
• IndexFlatL2 implementation with API parity to FAISS
• Documentation for client usage
• Drop-in replacement behavior for seamless FAISS integration
• Test suite for client functionality