RushDB is a unified memory layer for AI agents, consolidating graph, vector, and key-value stores into a single API. It allows pushing any JSON data and querying it via graph traversal and semantic search without schema design or migrations, simplifying AI development and data management.

How It Works

The system utilizes a Labeled Meta Property Graph (LMPG) model, elevating properties to first-class nodes. This approach automatically infers types, relationships, and ontologies on write, enabling a unified query surface that combines graph traversal and semantic search. This eliminates manual schema modeling and separate embedding pipelines, dynamically linking related data.

Quick Start & Requirements

RushDB offers a managed Cloud service (free tier available) or self-hosting via Docker.

• Install: npm install @rushdb/javascript-sdk or pip install rushdb.
• Self-host: Requires Docker and Neo4j (2026.01.4+ with APOC plugin recommended).
• Docs: https://docs.rushdb.com
• Examples: https://github.com/rush-db/examples

Highlighted Details

• Unified Data Layer: Integrates graph, vector, and key-value storage into one API.
• Zero Schema & Inferred Ontology: Accepts any JSON, automatically inferring types, relationships, and building a queryable ontology.
• Managed Embeddings: Server-side auto-embedding for semantic search.
• Graph + Vector Querying: Combines semantic similarity and relationship traversal in single queries.
• MCP-Native: Includes an MCP server for natural language interaction with memory and data.
• ACID Transactions: Ensures data consistency via Neo4j backend.

Maintenance & Community

• Website: https://rushdb.com
• Docs: https://docs.rushdb.com
• X Handle: @rushdb
• Development: Active CI, open to contributions.

Licensing & Compatibility

• Core Platform: Elastic License 2.0 (may restrict managed service offerings).
• SDKs, Docs, Website: Apache License 2.0 (generally permissive for commercial use).

Limitations & Caveats

RushDB is not suited for extreme write throughput demands or sub-millisecond latency requirements due to its Neo4j foundation. It is also not appropriate if only flat key-value storage is needed, or if a rigid, database-enforced relational schema is a strict requirement, as RushDB is schema-free by design.