Hexcore

Hexcore is a modular C++ vector database focused on clarity of architecture and extensibility.

Architecture Overview

Layer	Responsibility	Key Types
Core	Fundamental types, vector math, memory management	Vector, Distance, MemoryPool, types.hpp
Storage	Persistence + metadata indexing	StorageInterface, FileStorage, MemoryStorage, VectorStore, MetadataStore
Indexing	Approximate / exact nearest neighbor search	IndexInterface, FlatIndex (+ IVF/HNSW/PQ placeholders)
API	High-level user operations & orchestration	HexcoreClient, Database, Collection, configs & query builders

Features & Roadmap

Area	Item	Status
Core	Vector abstraction with dimension metadata; normalization utilities	[x]
Distance	L2, L2 Squared, Cosine, Inner Product, L1 via pluggable Distance strategies	[x]
Memory	Configurable MemoryPool and pool-backed temporary allocations	[x]
Storage	MemoryStorage (in‑process)	[x]
Storage	FileStorage (VectorStore + MetadataStore)	[x]
Storage	Tag/attribute inverted indexes	[x]
Indexing	Pluggable IndexInterface; baseline FlatIndex implemented	[x]
API	High-level HexcoreClient, Database, Collection; Config/ConfigBuilder	[x]
Query	QueryBuilder, QueryOptions, QueryFilter; selective vectors/metadata/distances; thresholds and exclusions	[x]
Observability	Collection stats (vector count, index built, storage usage)	[x]
Indexing & Search	IVF Flat (coarse quantization)	[ ]
Indexing & Search	IVF-PQ (product quantization)	[ ]
Indexing & Search	HNSW graph index	[ ]
Indexing & Search	PQ/OPQ training pipeline & codebook persistence	[ ]
Indexing & Search	Graph maintenance utilities (level promotion, lazy deletion)	[ ]
Query & Filtering	Boolean tag/attribute expressions (AND/OR/NOT DSL)	[ ]
Query & Filtering	Result caching & prefetch strategies	[ ]
Server & Interfaces	HTTP/REST service (collection management, insert, search)	[ ]
Server & Interfaces	gRPC API for high‑throughput streaming inserts	[ ]
Server & Interfaces	WebSocket streaming search / progress events	[ ]
Server & Interfaces	Observability: Prometheus metrics, structured logging, tracing (OTel)	[ ]
Performance & Systems	SIMD acceleration (AVX2/AVX512) & tuned fallbacks	[ ]
Performance & Systems	Thread pool / task scheduler for batch search & index builds	[ ]
Performance & Systems	Parallel index building & incremental rebuild triggers	[ ]
Developer Experience	Python bindings (Pybind11)	[ ]
Developer Experience	Benchmark suite & dataset loaders	[ ]

Build

Requirements: CMake >= 3.14, a C++17 compiler (GCC / Clang), Threads library.

git clone https://github.com/lalitx17/Hexcore.git
cd Hexcore
mkdir build && cd build
cmake ..
cmake --build . -j$(nproc)
./bin/examples/hexcore_example

References

Faiss

HNSW

Faiss Blog

How Vector DBs work