Open-source graph databases
Choose for the relationships you need to traverse.
Tracing account ownership, exploring service dependencies and retrieving connected evidence are graph-shaped problems. A graph database makes nodes, relationships and traversal central to the data model. That warrants its own comparison alongside OLAP and vector databases.
Sources reviewed: 2026-09-06. This is documentary evaluation from official repositories and manuals. Shortlists are editorial judgments, not benchmark scores. The selected versions identify the review scope, not a guarantee of maintenance or a recommendation to deploy without further evaluation.
At a glance
Rating (1–5): an editorial readiness score for the fit named in the same row, dated 2026-09-06. It adds five criteria worth 0, 0.5 or 1 point each: maintenance, open edition (what the open-source edition includes without a paid tier), maturity and community, operating scope and interoperability, with caps for archived, stale or prerelease projects. The breakdown sits under the table and the method on the Blog index. A rating is not a benchmark or a universal ranking; the workload sections still decide.
| Project | Rating | Reviewed core license | I would shortlist it for | Main constraint |
|---|---|---|---|---|
| Neo4j Community 2026.07.1 | GPLv3 | A dedicated property-graph server and Cypher application | Community is for a single instance; clustering and online backup belong to Enterprise. | |
| Apache AGE 1.8.0 for PostgreSQL 18 | Apache-2.0 | Graph queries alongside existing PostgreSQL data | PostgreSQL/extension compatibility and query semantics need an explicit matrix. | |
| JanusGraph 1.1.0 | Apache-2.0 code | Distributed property graphs with TinkerPop/Gremlin | Storage and optional mixed-index backends add independent operating responsibilities. | |
| LadybugDB 0.20.2 | MIT | Embedded graph analytics inside an application or local data workflow | Application ownership of concurrency, files and recovery differs from a managed graph server. | |
| NebulaGraph 3.8.0 | Apache-2.0 | A distributed graph service with separate query, metadata and storage roles | The reviewed release dates to 2024; assess maintenance, dependencies and operational support. | |
| OrientDB 3.2.56 | Apache-2.0 | Applications combining graph and document modeling | Its model, query language and deployment need a trial beyond a graph-only example. |
How each rating was computed
Five criteria worth 0, 0.5 or 1 point each. Caps: archived upstream 1, no stable release in 18 months 2, no general-availability release 2.5. "Open edition" scores what the open-source edition includes without a paid tier. Scored on 2026-09-06 from the official repository, releases and documentation; the method is on the Blog index.
| Project | Maintenance | Open edition | Maturity | Operations | Interoperability | Rating |
|---|---|---|---|---|---|---|
| Neo4j Community | 1 | 0.5 | 1 | 0.5 | 1 | 4 |
| Apache AGE | 0.5 | 1 | 0.5 | 1 | 0.5 | 3.5 |
| JanusGraph | 0 | 1 | 1 | 0.5 | 1 | 2 (capped from 3.5) |
| LadybugDB | 1 | 1 | 0.5 | 1 | 0.5 | 4 |
| NebulaGraph | 0 | 0.5 | 1 | 0.5 | 0.5 | 2 (capped from 2.5) |
| OrientDB | 1 | 1 | 0.5 | 1 | 0.5 | 4 |
Neo4j Community uses GPLv3, not AGPLv3. Its commercial editions and separately distributed tools have their own boundaries. A permissive or copyleft license does not itself establish scalability or availability.
Start with the workload
- An application built around Cypher traversals: start with Neo4j Community for a single-instance trial. If availability requirements need clustering, compare the appropriate licensed edition and distributed alternatives.
- A PostgreSQL application with graph-shaped queries: evaluate AGE before adding another persistent service. Include contention with the existing transactional workload and the extension upgrade path.
- Embedded analytical exploration: try LadybugDB, already present in this site's technology stack. Compare application integration and representative graph queries rather than server feature counts.
- A graph that requires distributed storage: compare JanusGraph and NebulaGraph with the intended topology. Budget their dependencies and establish an acceptable maintenance and recovery plan.
- Documents and relationships belong in one application model: include OrientDB. Verify that combining models simplifies the actual application.
For a small fixed hierarchy, first test a relational model and recursive queries on the database already operated. Adding a graph service should solve a demonstrated query or modeling problem.
Version and operating notes
Neo4j — the edition changes the availability design
The operations manual distinguishes single-instance Community from Enterprise clustering, failover and online backups. The code license covers the reviewed Community source.
Trial import, transactions, constraints and restoration with the actual edition. Check the Cypher version and driver combination. Including a capability from an Enterprise demonstration in a Community operating plan would leave a gap precisely when recovery is needed.
AGE — pin the extension to PostgreSQL
The project overview describes graph querying integrated with PostgreSQL. The download matrix identifies different AGE releases for different PostgreSQL majors. This review uses the 1.8.0 release for PostgreSQL 18, published on GitHub on 2026-07-09 under a tag that keeps an rc0 suffix; the download page still listed 1.7.0 at review time.
Test SQL/Cypher composition, data types, permissions and query plans on that pairing. Recover both relational and graph data, then rehearse the extension and PostgreSQL upgrade. Cypher syntax support is not automatic compatibility with every Neo4j procedure or query.
JanusGraph — the backend is part of the choice
JanusGraph uses the property-graph model through TinkerPop and Gremlin, with configurable storage and indexing backends. Its license file distinguishes Apache-licensed code from documentation licensing.
Choose the backend before evaluating consistency and failure behavior. Recover graph data and indexes together; test traversal results during reindexing. The reviewed 1.1.0 release is from 2024, so maintenance and supported dependency versions need explicit acceptance criteria.
LadybugDB — embedded deployment changes responsibility
The manual describes an embedded, columnar property-graph engine with Cypher and transactional support. Its process and data-file lifecycle belong to the application.
Test concurrent application requests, memory pressure and a process interruption during a write. Restore the files and reconcile edges as well as nodes. Evaluate the documented Cypher differences before treating an existing Neo4j application as portable.
NebulaGraph — query, metadata and storage fail differently
The 3.8 manual describes a distributed graph with nGQL and separate service roles. Its openCypher-related syntax needs feature-level migration checks.
Rehearse loss of a storage node and metadata service, and measure the effect of high-degree vertices. Verify the backup tooling for the exact deployment. Review release age and the dependency support path; an unarchived repository alone does not settle those questions.
OrientDB — validate the combined data model
The project combines document and graph models. Its 3.2 documentation covers the server and database operating model.
Trial a document update that changes relationships and a traversal that crosses document classes. Confirm transaction boundaries and restore behavior. Compare the final application model, not an isolated traversal against a product configured for a different workload.
Other names need a license check
The reviewed Memgraph license uses BSL and enterprise terms. FalkorDB uses SSPL, and the current ArangoDB license uses BSL. They should be evaluated under those exact terms, separately from the open-source shortlist above. Public source code and free downloads do not establish the same license rights.
Graphs, vectors and Graph RAG
A graph traversal answers questions about explicit relationships. Vector retrieval finds candidates by similarity. Some products offer both, but the query semantics and evaluation targets remain different.
For Graph RAG, test whether relationships improve retrieval on judged questions. Preserve the provenance of extracted facts, handle conflicting claims, and enforce permissions while expanding paths. A plausible generated relationship is not an authoritative fact. Compare against the same vector-retrieval baseline.
A trial that can change the decision
Use accounts, transactions and ownership relationships with skewed degrees.
- Correctness: include cycles, duplicate edges, deleted nodes and multi-step updates; check the expected paths and transaction outcomes.
- Queries: test bounded traversals, high-degree vertices and the actual filters. Record tail latency with concurrent writes and representative indexes.
- Authorization: prevent cross-tenant path expansion and inspect exports, procedures and direct query access.
- Recovery: restore nodes, relationships, indexes, schema and access settings, then reconcile against the source data.
- Exit: export identities and relationships, port representative queries and measure the effort required to preserve their meaning.
Follow the data
Orchestration coordinates ingestion; transformation normalizes entities and identifiers. Use observability to detect stale graph projections, and keep object storage for source and recovery artifacts.