NornicDB
A graph database that combines Neo4j-compatible Bolt and Cypher access with vector search and MVCC-based historical reads, aimed at agent memory, graph-RAG and audit-friendly knowledge stores.
NornicDB is a graph database built for workloads that need graph traversal, vector retrieval and historical truth in the same system rather than a vector store bolted onto a separate graph engine. It speaks Neo4j's Bolt and Cypher protocol for compatibility with existing drivers, and also exposes REST, GraphQL and gRPC interfaces, with an option to preserve Qdrant-style client workflows for migration.
Its design draws on research in temporal GraphRAG, agent memory systems and event-sourced decision memory, which shows up as graph-native temporal facts, versioned canonical knowledge, replayable history and policy-driven memory retention with sub-millisecond HNSW search. It supports CPU, CUDA, Metal and Vulkan backends, runs through Docker images, and ships agent-ready skill files describing its Cypher query surface for use with Claude and other agents.