NornicDB

概述

NornicDB 是一个高性能的图和向量数据库，结合了时间 MVCC 和亚毫秒级 HNSW 搜索功能。它提供 Neo4j Bolt/Cypher 兼容性和 Qdrant gRPC 支持，使现有应用易于迁移。数据库包含智能功能，如模式、托管嵌入、LLM 重排序、GPU 加速和集成的 MCP 服务器，可增强 AI 工作流程。在基准测试中，它比 Neo4j 实现了 12x-52x 的性能提升，同时在单个引擎中提供混合图+向量检索能力。

试试问 AI

装完之后，这里有 3 个你可以让 AI 做的事：

• query_neo4j

  Execute Cypher queries on the graph database

  试试:Find all nodes with label Person试试:Create a new relationship between two nodes
• vector_search

  Perform vector similarity search using HNSW algorithm

  试试:Find similar documents to the one about machine learning试试:Search for products with vector embeddings
• graph_traversal

  Traverse the graph structure with path-finding capabilities

  试试:Find the shortest path between two nodes试试:Traverse all nodes connected to a starting node
• create_node

  Create a new node in the graph database

  试试:Create a new User node with email property试试:Add a new Product node with price and description
• create_relationship

  Create a new relationship between two nodes

  试试:Connect a User node to an Order node试试:Create a LIKES relationship between a Person and a Movie
• update_node

  Update properties of an existing node

  试试:Update the name property of a Person node试试:Change the status of an Order node
• delete_node

  Delete a node and its relationships from the graph

  试试:Remove a User node and all connected relationships试试:Delete an obsolete Product node
• hybrid_retrieval

  Perform combined vector search and graph traversal in one query

  试试:Find similar documents and their connected concepts试试:Search for products and their related reviews
• create_vector_index

  Create a new vector index for similarity search

  试试:Create an index for document embeddings试试:Set up a vector index for product features
• historical_read

  Query past versions of the graph using MVCC

  试试:Check how the graph looked last week试试:Find previous values of a node property
• schema_constraint

  Add or remove constraints on node properties

  试试:Create a unique constraint for email properties试试:Add a NOT NULL constraint for name properties
• graph_rollback

  Rollback the graph to a previous transaction state

  试试:Rollback to the state before the last batch of updates试试:Restore graph from yesterday's backup

说明：Tool names inferred from documented database capabilities and common graph database operations. The README mentions Neo4j compatibility and vector search but doesn't provide explicit MCP tool names.

可对比工具

安装

# Apple Silicon
docker run -d --name nornicdb -p 7474:7474 -p 7687:7687 -v nornicdb-data:/data timothyswt/nornicdb-arm64-metal-bge:latest

# AMD64 / 仅 CPU
docker run -d --name nornicdb -p 7474:7474 -p 7687:7687 -v nornicdb-data:/data timothyswt/nornicdb-amd64-cpu-bge:latest

git clone https://github.com/orneryd/NornicDB.git
cd NornicDB
go build -o nornicdb ./cmd/nornicdb
./nornicdb serve

{
  "mcpServers": {
    "nornicdb": {
      "command": "nornicdb",
      "args": ["mcp"]
    }
  }
}

Hacker News 讨论

开发者社区最近的相关讨论。

• Cutting Cypher Query Latency: Streaming Traversal and Query Shape Specialization▲ 1
  帖子 by orneryd · 2026-02-26

NornicDB 对比