Neo4j

Graph database with Cypher query language plus vector search.

Reviewed by 7wData
Updated

On this page

Publisher review

Neo4j is a property graph database designed for storing and querying highly connected data. It represents entities as nodes and their relationships as directed, typed edges, allowing it to model complex networks more intuitively than relational or document-oriented databases.

The core strength is Neo4j's index-free adjacency architecture: each node directly references its neighbors, making traversal time constant regardless of total graph size. This enables efficient queries across multiple relationship hops—a workload where Neo4j outperforms PostgreSQL or MongoDB by orders of magnitude on specific patterns. Cypher, the native query language, is deliberately readable and closer to natural language than SQL, making complex relationship queries more intuitive to reason about.

Neo4j ships in three forms: a free Community Edition (GPL3-licensed), fully-managed AuraDB ($0 to $146/GB/month depending on tier), and Enterprise/Infinigraph editions for self-hosted or hybrid deployments. Aura's pricing structure is competitive for small to mid-sized graphs (up to 32GB), but grows expensive at scale; typical production instances cost $800–$3,000 monthly.

In 2025–2026, Neo4j has aggressively integrated generative AI patterns. Vector search (HNSW-backed) is now native, supporting hybrid semantic+structural queries. Official integrations span LangChain, LlamaIndex, Spring AI, and the new GraphRAG Python package—making it a natural choice for knowledge graph applications or retrieval-augmented generation systems. Bloom provides non-technical graph exploration and visualization.

The weaknesses are tangible. Cypher has a steep learning curve for newcomers unfamiliar with graph concepts; Text-to-Cypher LLM translation still struggles with query optimization and aggregation choices. Performance benchmarks show TigerGraph outpacing Neo4j 12–58× on data loading and 40–337× on certain read patterns, though this depends heavily on query shape. The community is small compared to PostgreSQL or MongoDB, limiting third-party tooling and operational expertise.

Neo4j is best suited for organizations whose core problem domain is fundamentally graph-shaped: social networks, fraud detection rings (suspicious transaction webs), authorization systems, or knowledge graphs. For teams uncertain whether their use case is truly graph-centric, PostgreSQL with JSON or MongoDB often suffices with lower operational friction.

Get the AI & data signal, daily.

335k+ subscribers read this every morning. One email, both newsletters. Unsubscribe anytime.

How it works

  1. Cypher Query Language

    Declarative graph query language optimized for relationship traversal; uses pattern matching to find connected nodes and edges in readable, SQL-like syntax.

  2. Vector Search (HNSW)

    Native approximate nearest neighbor search for embedding-based retrieval; enables hybrid queries combining semantic similarity with structural graph traversal.

  3. Graph Data Science Library

    70+ graph algorithms for centrality, community detection, similarity, and machine learning inference; runs algorithms directly on the database without data export.

  4. Neo4j Bloom

    Visual graph exploration tool with perspective-based views; allows non-technical users to explore relationships and create saved queries without writing Cypher.

  5. ACID Transactions

    Full transaction support with isolation levels and rollback capabilities; ensures data consistency for production-critical applications.

  6. GenAI Framework Integrations

    Official packages for LangChain, LlamaIndex, and GraphRAG patterns; supports embedding generation via OpenAI, Google Vertex, Azure, and AWS Bedrock.

  7. Hybrid Search

    Combines vector similarity search with structural graph queries in a single Cypher operation; enables knowledge graph retrieval with both semantic and relationship filters.

Strengths and trade-offs

Strengths

  • Index-free adjacency enables constant-time graph traversal regardless of graph size; outperforms relational databases 10–100× on multi-hop relationship queries.
  • Cypher is readable and intuitive for relationship-centric queries compared to SQL JOINs or NoSQL aggregation pipelines; reducing query complexity and development time.
  • Native vector search plus official GenAI framework integrations (LangChain, LlamaIndex, GraphRAG) position it as a platform for knowledge graphs and retrieval-augmented generation, not just a database.

Trade-offs

  • Pricing scales aggressively ($65–146/GB/month on Aura); graphs larger than 32GB quickly exceed relational or warehouse cost profiles for comparable data volumes.
  • Steep learning curve: Cypher requires schema familiarity and graph-query intuition; Text-to-Cypher LLM translation still struggles with aggregation and join optimization decisions.
  • Performance benchmarks show TigerGraph 12–58× faster on data loading and 40–337× faster on certain read patterns; smaller community than PostgreSQL or MongoDB limits third-party tooling and operational expertise.

Pricing context

Neo4j offers free Community Edition (GPL3-licensed) and three AuraDB tiers: Free ($0, learning only), Professional ($65/GB/month, 1–128GB), and Business Critical ($146/GB/month, 2–512GB with 99.95% SLA). Typical production instances cost $800–$3,000/month for 16–32GB graphs. Enterprise and Infinigraph editions (unlimited horizontal scaling) require sales negotiation. Unlike data warehouses, Aura bundles storage, I/O, backups, and data transfer in the monthly rate, avoiding surprise overage charges but locking pricing to data volume.

Getting started with Neo4j

  1. Sign up for Neo4j

    Choose between Community Edition (free, self-hosted), Aura Free tier (cloud, learning only), or request an Enterprise trial. Create an account at neo4j.com for managed Aura, or download Community locally. Aura Free is best for evaluation; both require internet access for initial setup.

  2. Load your data

    Use Cypher CREATE statements to define nodes and relationships, or import CSV files via Neo4j's bulk loader. Map your entities to nodes and connections to typed, directed edges. Start with small datasets to validate your schema before scaling.

  3. Define relationships with Cypher

    Learn Cypher syntax and write exploratory queries to traverse your data model. Use pattern matching notation (e.g., (person)-->(company)) to fetch connected nodes. Cypher is readable compared to SQL, but requires graph-thinking; start with simple two-hop relationship queries.

  4. Run your first query

    Execute a query that traverses multiple relationship hops to validate your model. Optionally run a Graph Data Science algorithm (centrality, community detection) to surface insights without exporting data. Review results to confirm your schema and reveal structural patterns.

  5. Enable vector search and backups

    Enable automated backups on Aura Professional for production use. Enable vector search (native HNSW) if your application requires semantic similarity queries. Connect LangChain or LlamaIndex for generative AI integration, or upgrade your Aura tier as data volume increases.

Frequently Asked Questions

What is Neo4j?

Neo4j is a property graph database that stores data as nodes (entities) and edges (relationships). It's designed for querying highly connected data using an index-free adjacency architecture, where each node directly references its neighbors. This enables constant-time traversal regardless of graph size, outperforming traditional relational databases on multi-hop relationship queries.

How does Neo4j's Cypher language work?

Cypher is Neo4j's declarative query language designed for relationship-centric queries. It uses pattern matching to find connected nodes and edges in readable, SQL-like syntax. Cypher is more intuitive than SQL JOINs for complex relationships, reducing query complexity and development time. It's closer to natural language than traditional database query languages.

What are the main pricing tiers for Neo4j?

Neo4j offers a free Community Edition and three AuraDB cloud tiers. AuraDB includes Free ($0, learning only), Professional ($65/GB/month, 1–128GB), and Business Critical ($146/GB/month, 2–512GB with 99.95% SLA). Typical production instances cost $800–$3,000 monthly for 16–32GB graphs. Monthly pricing bundles storage, I/O, backups, and transfers.

Does Neo4j support vector search and AI integration?

Yes. Neo4j includes native vector search using HNSW (approximate nearest neighbor) for embedding-based retrieval. It supports hybrid queries combining semantic similarity with structural graph traversal. Official integrations include LangChain, LlamaIndex, GraphRAG, and major LLM providers: OpenAI, Google Vertex, Azure, and AWS Bedrock. This positions Neo4j as a knowledge graph platform.

How does Neo4j compare to PostgreSQL?

Neo4j outperforms PostgreSQL 10–100× on multi-hop relationship queries due to index-free adjacency architecture. PostgreSQL requires complex JOINs for relationships; Neo4j traverses directly. However, for non-graph workloads, PostgreSQL often suffices with lower operational friction. Neo4j shines for fundamentally graph-shaped problems: social networks, fraud detection, authorization systems, and knowledge graphs.

What are Neo4j's main weaknesses?

Cypher has a steep learning curve for graph-query newcomers. Pricing scales aggressively ($65–146/GB/month on Aura), making graphs larger than 32GB expensive versus warehouses. TigerGraph outpaces Neo4j 12–58× on data loading and 40–337× on certain read patterns. Neo4j's community is smaller than PostgreSQL or MongoDB, limiting third-party tooling and operational expertise.

Alternatives in this category

Integrations

LangChain LlamaIndex Spark

How Neo4j compares

Direct head-to-head against 3 competitors. Picked by 7wData.

This tool

Neo4j

Pricing
Neo4j offers free Community Edition (GPL3-licensed) and three AuraDB tiers: Free ($0, learning only), Professional ($65/GB/month, 1–128GB), and Business Critical ($146/GB/month, 2–512GB with 99.95% SLA). Typical production instances cost $800–$3,000/month for 16–32GB graphs. Enterprise and Infinigraph editions (unlimited horizontal scaling) require sales negotiation. Unlike data warehouses, Aura bundles storage, I/O, backups, and data transfer in the monthly rate, avoiding surprise overage charges but locking pricing to data volume.
Target
Neo4j is a property graph database designed for storing and querying highly connected data.
Deployment
hybrid
Strength
Index-free adjacency enables constant-time graph traversal regardless of graph size; outperforms relational databases 10–100× on multi-hop relationship queries.
Watch for
Pricing scales aggressively ($65–146/GB/month on Aura); graphs larger than 32GB quickly exceed relational or warehouse cost profiles for comparable data volumes.

TigerGraph

Pricing
Savanna cloud: $0.025/GB/month storage; compute metered hourly by workspace size. Enterprise on-prem requires custom quote.
Target
Enterprise analytics teams running deep multi-hop traversal workloads across billions of edges: fraud detection, supply chain, recommendation.
Deployment
SaaS (Savanna), on-prem, hybrid
Strength
Massively parallel GSQL processing benchmarked 12-58x faster than Neo4j on data loading and 40-337x on certain read patterns.
Watch for
Schema changes require query recompilation and often data reloading; distributed cluster management demands specialized expertise not required by Neo4j.

Amazon Neptune

Pricing
On-demand db.r6g.large ~$0.34/hr; storage $0.10/GB/month; I/O $0.22/million requests (Standard tier). No minimum commitment.
Target
AWS-native engineering teams building connected-data workloads inside existing AWS infrastructure without self-hosting database overhead.
Deployment
Fully managed SaaS on AWS
Strength
Supports both RDF/SPARQL and property graph (Gremlin, openCypher) in one managed service with native AWS IAM and VPC controls.
Watch for
Deep AWS lock-in; no native data science algorithm library comparable to Neo4j GDS; I/O costs escalate without the I/O-Optimized tier.

ArangoDB

Pricing
Community Edition: BSL 1.1 (Apache 2 pre-v3.12), 100GB single-cluster cap. Oasis managed cloud from $0.20/hr. Enterprise: custom quote.
Target
Teams wanting document, key-value, and property graph workloads on one database to avoid separate purpose-built stores.
Deployment
Open-source, SaaS (Oasis), on-prem, hybrid
Strength
Multi-model design runs document, key-value, and property graph workloads under one query language (AQL), eliminating polyglot database overhead.
Watch for
2023 re-license from Apache 2 to BSL 1.1 for v3.12+; Community Edition now imposes a 100GB cap on single-cluster dataset size.

User reviews

No user reviews yet. Be the first to write one.

Sources

Reporting on this tool draws on these publicly available sources.

  1. neo4j.com — AuraDB pricing tiers (Free, Professional, Business Critical) and cost structure ($65–146/GB/month)
  2. neo4j.com — Company founding year (2007), founders (Emil Eifrem, Johan Svensson, Peter Neubauer), and headquarters (San Mateo, California; Silicon Valley move in 2011)
  3. risingwave.com — Comparative performance: TigerGraph 12–58× faster on data loading, strengths in read-heavy workloads for Neo4j, and best-use-case guidance
  4. neo4j.com — GenAI integrations (LangChain, LlamaIndex, GraphRAG), vector search capabilities, and supported LLM providers (OpenAI, Azure, Google Vertex, AWS Bedrock)