What it is
Neo4j stores data as nodes and relationships and queries it with Cypher, a pattern-matching language. The GitHub repository holds the Community Edition, written in Java and licensed under GPLv3. It includes the graph engine, ACID transactions, Cypher, full-text indexes and vector indexes.
For AI work, Neo4j 2025.10 added a native VECTOR property type, and since 2026.02 the Cypher SEARCH clause can filter inside a vector index, so a vector query restricted by graph properties still returns the number of rows asked for. This works in Community, Enterprise and Aura. Neo4j also maintains a GraphRAG Python library, and most graph RAG and agent-memory projects, including Graphiti and LightRAG, can use it as their graph store.
Community Edition runs on a single server. Clustering, hot backups, multiple databases and role-based security require the commercial Enterprise Edition or AuraDB, the managed cloud service.
Our review
Neo4j is the safe choice of graph database for AI work, mainly because everything else supports it. Graphiti recommends it for new projects, LightRAG lists it as a graph backend, and Neo4j's own GraphRAG library and Aura Agent build retrieval agents on top. The repository is active, with about 17,350 GitHub stars and Community Edition 2026.10.1 tagged on 20 October 2026; releases appear as tags, not GitHub releases.
Its vector support is now credible rather than bolted on. In-index filtering through the Cypher SEARCH clause, generally available since 2026.02 and included in Community Edition, fixes the classic post-filtering problem where a filtered vector query returns too few results. That means one database can hold both the graph and the embeddings for a modest RAG system.
The open-source edition is meaningful but limited. Community Edition is GPLv3 and single-server, with no clustering, hot backups or fine-grained access control. Both editions also send anonymous usage data to Neo4j by default (dbms.usage_report.enabled is true), which Neo4j's own security benchmark recommends switching off. Production deployments therefore tend to end up on Enterprise, which has no published price, or AuraDB, where Business Critical costs from $0.20 per GB of memory per hour with a 99.95% SLA.
A graph helps RAG when questions follow relationships: who reports to whom, which contracts reference which suppliers, what changed and when. For semantic lookup over documents, pgvector or Qdrant is simpler and cheaper, and a 16 GB Business Critical instance left running all month costs roughly $2,300, before anyone has asked a question that needed a graph.
Where it earns its place
A legal-tech team building contract search
Answer questions like which suppliers are bound by clauses that reference a sanctioned entity
Store contracts, clauses and parties as nodes with embeddings on clause text, then filter vector search by graph properties with the Cypher SEARCH clause. Community Edition is free on one server; turn off usage reporting before loading client data.
An AI engineer adding long-term memory to a support agent with Graphiti
Keep a time-aware record of what each customer said and changed
Run Neo4j Community in Docker as Graphiti's backend, or AuraDB Professional from $0.09 per GB-hour (1 GB minimum, about $66 a month) to avoid running a database. The graph lets the agent answer "what changed since last month".
A CTO at a 200-person fintech budgeting for production
Decide between self-managed Enterprise and managed AuraDB
AuraDB Business Critical starts at $0.20 per GB-hour with a 2 GB minimum, multi-zone and a 99.95% SLA, and pausing cuts the bill by 80%. Enterprise for on-prem clustering is sales-quoted, so an Aura estimate is the only published benchmark.
Before you commit
- Setup
- Docker image, Neo4j Desktop or a tarball for Community; a first database runs on localhost:7474 (browser) and 7687 (Bolt) within minutes. AuraDB Free needs no card.
- Integrations
- Official drivers for Python, Java, JavaScript, .NET and Go, the neo4j-graphrag Python library, LangChain and LlamaIndex integrations, and backends for Graphiti, LightRAG and Graphify.
- Data and privacy
- GPLv3 Community; closed-source Enterprise needs a commercial licence. Anonymous usage data goes to udc.neo4j.com by default; set dbms.usage_report.enabled=false. AuraDB compliance details sit in the Trust Center at trust.neo4j.com rather than on the pricing page.
- Cost at scale
- Self-hosted Community costs your server, typically a memory-heavy VM. AuraDB Professional from $0.09/GB/hour (up to 128 GB), Business Critical from $0.20/GB/hour (up to 2 TB on GCP); storage, IO and backups are included.
- Limits
- Community is single-server, with no clustering, hot backups or fine-grained access control; the free Community Edition of the Graph Data Science library caps algorithm concurrency at four CPU cores and the model catalogue at three models. About 233 open issues and pull requests on GitHub, where the issue tracker is a minor support channel.
What's good
- Supported by most graph RAG and agent-memory tools
- Vector indexes with in-index filtering in the free edition
- Cypher is expressive for relationship queries
- AuraDB pricing includes storage, IO and backups
What's not
- Community Edition is single-server with no clustering or hot backups
- Usage reporting to Neo4j on by default
- GPLv3 may not suit some commercial embedding
- Enterprise self-managed pricing not published
Best for
Teams whose RAG or agent-memory questions depend on relationships between entities, and who need a proven graph store.
Alternatives
Neo4j: common questions
- Is Neo4j worth it?
- The default graph store when your RAG or agent memory genuinely needs relationships, but the free edition stops at one server and most RAG does not need a graph at all. We rate it 4.1 out of 5.
- How much does Neo4j cost?
- Community Edition free under GPLv3. Enterprise by quote. AuraDB: Free tier; Professional $0.09/GB/hour (1 GB minimum); Business Critical $0.20/GB/hour (2 GB minimum).
- What are the downsides of Neo4j?
- Community Edition is single-server with no clustering or hot backups; usage reporting to Neo4j on by default; gPLv3 may not suit some commercial embedding; enterprise self-managed pricing not published.
- Who is Neo4j best for?
- Teams whose RAG or agent-memory questions depend on relationships between entities, and who need a proven graph store.
- How much does Neo4j cost for a team or in production?
- Self-hosted Community costs your server, typically a memory-heavy VM. AuraDB Professional from $0.09/GB/hour (up to 128 GB), Business Critical from $0.20/GB/hour (up to 2 TB on GCP); storage, IO and backups are included.
- Is Neo4j safe for company data?
- GPLv3 Community; closed-source Enterprise needs a commercial licence. Anonymous usage data goes to udc.neo4j.com by default; set dbms.usage_report.enabled=false. AuraDB compliance details sit in the Trust Center at trust.neo4j.com rather than on the pricing page.
- What are the best Neo4j alternatives?
- The closest alternatives we have reviewed are Graphiti, LightRAG, GraphRAG and pgvector.
Answers checked Oct 23, 2026.