orneryd/NornicDB

Built on an MVCC snapshot isolation model, NornicDB unifies graph traversal, vector search, and temporal/audit queries in a single execution engine rather than bolting components together—hybrid retrieval patterns (vector search followed by graph expansion) execute in one path without cross-system coordination. The architecture uses shape-specialized streaming executors for common patterns while maintaining a general Cypher fallback, with configurable parser modes (custom low-overhead vs. ANTLR strict) and hardware acceleration via Metal/CUDA/Vulkan for embedding operations. It targets knowledge systems, Graph-RAG, agent memory layers, and canonical truth stores where semantic search alone is insufficient—workloads requiring as-of reads, tritemporal facts, transaction conflict detection, and historical MVCC retention alongside vector retrieval.

267 stars.