orneryd/NornicDB

Nornicdb is a distributed low-latency, Graph+Vector, Temporal MVCC with all sub-ms HNSW search, graph traversal, and writes. Using Neo4j Bolt/Cypher and qdrant's gRPC means you can switch with no changes while adding intelligent features like schemas, managed embeddings, reranking+llm, GPU accel, Auto-TLP, Policy-based Memory Decay, and MCP server.

What it solves

NornicDB is a unified database engine designed for AI workloads that require simultaneous access to graph traversals, vector retrieval, and historical truth (temporal data). It eliminates the need to maintain separate databases for graph storage (like Neo4j) and vector stores (like Qdrant), consolidating them into a single execution path to reduce latency and architectural complexity.

How it works

NornicDB implements a hybrid engine that supports Neo4j's Bolt protocol and Cypher query language, making it a drop-in replacement for many Neo4j workflows. It uses Snapshot Isolation and Multi-Version Concurrency Control (MVCC) to provide historical "as-of" reads and a canonical graph ledger for auditability. For performance, it utilizes hardware-accelerated pathways (CUDA, Metal, Vulkan) and specialized "hot-path" streaming executors for common query shapes. It also integrates native vector search with HNSW indexing and supports hybrid retrieval (combining vector search with graph expansion).

Who it’s for

It is built for developers creating knowledge systems, AI agent memory architectures, and Graph-RAG pipelines who need a single system for task tracking, dependency graphs, and semantic retrieval.

Highlights

  • Neo4j & Qdrant Compatibility: Supports Bolt, Cypher, and Qdrant-compatible gRPC interfaces.
  • Knowledge-Layer Scoring: Implements profile-driven decay and promotion scoring to model how information is forgotten or reinforced over time.
  • Hardware Acceleration: Native support for GPU acceleration via Metal, CUDA, and Vulkan.
  • Temporal Truth: Built-in support for tritemporal facts and historical reads via MVCC.
  • High Performance: Significant latency reductions compared to Neo4j in specific benchmarks (e.g., Northwind benchmark).
  • Auto-Relationships: Automatically creates connections based on embedding similarity, co-access patterns, and temporal proximity.

Related

  • Project
    jeffhajewski/latticedb

    LatticeDB is an embedded, single‑file property‑graph database that also provides native HNSW vector similarity search and BM25 full‑text indexing. It offers a unified Cypher‑style query language, ACID transactions, and bindings for Python, TypeScript/Node, Go, and Java, making it suitable for local knowledge‑graph apps, agent memory, and RAG pipelines.

  • Dispatch
    LatticeDB: An Embedded Single-File Graph Database with Vector and Full-Text Search

    LatticeDB is an embedded, single-file property-graph database that integrates native vector similarity search and BM25 full-text indexing into a single query layer for local-first AI and RAG applications.

  • Project
    nuclia/nucliadb

    NucliaDB is a hybrid AI search database that combines vector, full-text, and graph indexes to store and search unstructured data.

  • Project
    rush-db/rushdb

    A memory layer for AI agents that combines key-value, vector, and graph databases into a single schema-free API with automatic embedding and relationship inference.

  • Project
    paradedb/paradedb

    ParadeDB is a PostgreSQL extension that adds fast full‑text, BM25, and vector similarity search (including hybrid queries) plus filtering, faceting, and OLAP‑style aggregates—all inside a single SQL‑compatible database. Built with Rust (pgrx, Tantivy, DataFusion), it can be run locally via Docker or as a managed cloud service, and integrates with popular ORMs, AI agent tools, and PaaS platforms.