rush-db/rushdb

What it solves

RushDB is a memory layer designed for AI agents and applications that eliminates the need to manage multiple separate databases (such as Redis for key-value storage, Pinecone for vector search, and Neo4j for graph relationships). It provides a single API to handle structured data, semantic search, and complex relationships without requiring a predefined schema or manual embedding pipelines.

How it works

RushDB uses a Labeled Meta Property Graph (LMPG) model where properties are treated as first-class graph nodes. This allows the system to automatically infer schemas, types, and relationships from JSON data as it is pushed to the database. It manages embeddings server-side, meaning users can index a string property once and every subsequent write is automatically embedded. It is built on top of Neo4j to ensure ACID transactions and can be deployed via a managed cloud service or self-hosted via Docker.

Who it’s for

It is primarily for developers building AI agents and RAG (Retrieval-Augmented Generation) applications that need persistent, structured memory and the ability to perform hybrid queries (combining semantic search and graph traversal).

Highlights

• Zero Schema: Automatically infers labels, types, and relationships from any JSON shape upon write.
• Hybrid Querying: Combines semantic similarity and relationship traversal in a single API call.
• Managed Embeddings: Server-side embedding of string properties removes the need for a separate embedding pipeline.
• MCP-Native: Includes a Model Context Protocol (MCP) server, allowing agents like Claude or Cursor to introspect and query memory using natural language.
• Analytical Capabilities: Supports aggregations (sum, avg, count, min, max) and grouping for complex data analysis.
• Flexible Deployment: Available as a managed cloud service or self-hosted with Docker and Neo4j.