[Guide]

Best open source graph databases in 2026

By the Omnigraph team · Last checked

Twelve graph databases ship their engine under an OSI-approved license and are actively maintained. Neo4j Community Edition is still the default for Cypher and a large ecosystem. Dgraph, NebulaGraph and JanusGraph scale out across machines, LadybugDB embeds in your process, and Omnigraph and TerminusDB are the two that let you branch, merge and read the graph as it was.

Six more engines that often appear on “open source” lists (Memgraph, FalkorDB, ArangoDB, SurrealDB, Fluree, HydraDB) are source-available, not open source: their licenses restrict offering the database as a service to others.

We build Omnigraph, one of the engines on this list. Every claim here links to a public source, the comparison is a script you can rerun, and Omnigraph’s weak spots are listed with its strengths.

[Picks]

Best for

Branches, merges and reads at an earlier commit

Omnigraph, TerminusDB

Cypher, the largest ecosystem, the most tutorials

Neo4j Community Edition

An embedded graph in your process

LadybugDB

Scale-out across many machines

Dgraph, NebulaGraph, JanusGraph

Strict typing and rich modeling

TypeDB

A graph inside the PostgreSQL you run

Apache AGE

Graph plus documents in one engine

ArcadeDB, OrientDB

A graph on S3-compatible object storage

Omnigraph, HelixDB

[Method]

How we chose and checked

Inclusion rules

  1. The engine itself is under an OSI-approved license, read from the repo’s LICENSE file. A permissive SDK on top of a source-available server doesn’t count.
  2. It’s a database you run, not a query layer over another store or a framework (so PuppyGraph and Apache TinkerPop are out).
  3. It’s active: a release or commits in the last 12 months. That drops Cayley (Apache-2.0, last release 2019). Kùzu was archived in October 2025; its maintained fork, LadybugDB, is on the list.

OSI-approved means on the Open Source Initiative’s license list.

The runnable test

We loaded the same small dataset into all twelve engines and ran the same four steps, each in a fresh Docker container (LadybugDB as a pinned Python package), with the version pinned and logged.

  1. Create the schema each engine’s docs recommend for typed data, then load 12 people, 3 companies, 5 projects and the edges between them.
  2. Run a two-hop query: which projects do Acme’s employees contribute to? The script checks the answer against the dataset (Atlas, Beacon, Comet).
  3. Try two bad writes through the normal write path: a person with age: "forty", then one with an undeclared nickname. Record whether each is rejected, by default and again in the engine’s strict mode where it documents one.
  4. Where the engine documents branches or versioned reads: branch, write on the branch, check main is untouched, merge, then read the commit from before the merge.

Scripts, dataset and full logs: ModernRelay/graph-database-comparison. Rerun it with python run_all.py.

Sources

Every cell in the comparison table links to a public source (docs, release notes or code) and carries the date it was last checked.

Keeping it current

A script checks every repo’s license file, archived flag, latest release and activity on the first Monday of each month. When something moves, the table changes and the changelog says what changed.

[Run]

What the run showed

Every engine answered the two-hop query correctly. The differences show up when a write breaks the schema and when you want a branch.

Comparison run results, 2026-11-11
Engine (version run)Two-hop queryRejects age: "forty"Rejects undeclared nicknameStrict modeBranch, merge, older commit
Omnigraph v0.15.0passrejectedrejectednot neededpass, pass, pass
TerminusDB v12.0.9passrejectedrejectednot neededpass, pass, pass
Neo4j CE 2026.10.2passacceptedacceptedrefused in Community Editionnot available
Dgraph v25.5.1passrejectedacceptedrejects both (--limit mutations=strict)not available
NebulaGraph 3.8.0passrejectedrejectednot needednot available
HelixDB server v0.0.15passacceptedacceptednone documentednot available
JanusGraph 1.2.0passrejectedacceptedrejects both (schema.default=none, schema.constraints=true)not available
Apache AGE 1.8.0 (PG 18)passacceptedacceptednone documentednot available
TypeDB CE 3.14.2passrejectedrejectednot needednot available
LadybugDB 0.22.3passrejectedrejectednot needednot available
ArcadeDB 26.11.1passrejectedacceptednone documentednot available
OrientDB 3.2.58passrejectedacceptedrejects both (STRICTMODE TRUE on the class)not available

Every command and its output is in each engine’s run log in ModernRelay/graph-database-comparison.

“Not available” means the engine documents no branches or versioned reads, so there was nothing to run; the comparison table shows what each offers instead (usually backups you restore, not query).

Accepting undeclared properties lets you change the data model without a migration, and moves validation into your application. The test shows which engines reject a bad write without extra setup.

[Compare]

Open source graph database comparison

Checked November 11, 2026. Stars from GitHub on the same day.

✓ yes · ◐ partial · – no. “Hybrid search ✓” means graph, vector and full-text search in one query with built-in rank fusion; ◐ means some of those in one query, without fusion. Hover a mark for its note; each links to its source.

[Engines]

The engines

Omnigraph

MIT · Rust · 2.0k stars · Run log (v0.15.0)

What it is
A typed property-graph database in Rust, first released in April 2026. Schemas are declared in .pg files and queries are named, typed .gq files. The graph is stored as Lance datasets on local disk or S3-compatible storage, GCS or R2 (Azure in preview).
Best for
Agent and team workflows where changes should be reviewable: a branch per agent or task, three-way merge across the whole graph, and queries against any earlier commit with --snapshot. Graph, vector and full-text search combine in one query with Reciprocal Rank Fusion, and omnigraph-server has an MCP endpoint built in.
Watch out for
No horizontal scaling yet: one server holds the writer lease, and clients write through it on isolated branches. No LangChain or LlamaIndex integration (MCP, a TypeScript SDK and HTTP only). It’s the youngest project here and has the fewest stars. Omnigraph Cloud is in private preview.
Data modelproperty graph · Typed nodes and edges declared in `.pg` schemas
Query languagesGQ (own) · Named, typed `.gq` queries and mutations
Schema on writeyes · Engine enforces types, keys and constraints on every write
ACID transactionsyes · One mutation query is one atomic graph commit; multi-command work staged on a branch and merged atomically
Storagelocal disk, object storage · S3-compatible, GCS, R2; Azure in preview
Scale-outno · Read replicas and overlapping writer replicas not a supported topology yet
Vector searchyes · `nearest()` over indexed `Vector(N)` properties
Managed cloudpartial · Omnigraph Cloud in private preview, waitlist
+ All 24 sourced facts
LicenseMIT
Written inRust
First release2026 · v0.1.0 on 2026-04-11
Schema migrationsyes · `schema plan` previews, `schema apply` applies; renames keep identity via `@rename_from`
Concurrent writerspartial · One server owns the storage writer lease; clients write concurrently through it, isolated on branches
Branch and mergeyes · Branch per agent or task, three-way merge across the whole graph
Versioned readsyes · Query any earlier graph commit with `--snapshot`
Object storageyes · Graph lives as Lance datasets on S3-compatible, GCS or R2; Azure in preview
Larger than RAMyes · Graph lives in Lance datasets on disk or object storage, not in memory
Full-text searchyes · `search`, `fuzzy`, `match_text`, `bm25`
Hybrid searchyes · Graph, vector and full-text with Reciprocal Rank Fusion in one query
Embeddedyes · Rust engine crate; CLI runs in-process against local `./graph.omni` targets
Serveryes · `omnigraph-server`, HTTP and OpenAPI
MCP serveryes · Built into `omnigraph-server` at `/mcp`
Agent frameworksno · MCP, TypeScript SDK and HTTP; no LangChain or LlamaIndex integration
Runs without the vendoryes · MIT, self-hosted binaries, no licence key or control plane

TerminusDB

Apache-2.0 · Prolog, Rust · 3.5k stars · Run log (v12.0.9)

What it is
A document-graph database that stores typed JSON documents as an RDF graph under a closed-world schema. The engine core is SWI-Prolog, with the storage layer in Rust. Every write is an immutable commit.
Best for
Versioned data with a schema: branches, field-level three-way merge, diff, and queries at any commit. It has the broadest git-style feature set on this list (rebase, squash, push, pull and clone as well), and a migration API that renames fields and casts types.
Watch out for
The docs call it an in-memory database: data persists on disk, but queries run in memory, so the working set has to fit. Single node (clustering is in the commercial Enterprise edition). Vector, full-text and hybrid search and an MCP server arrive in the 12.1 release candidate, not the current stable release.
Data modeldocument graph · JSON and JSON-LD documents stored as RDF triples under a schema
Query languagesWOQL, GraphQL · WOQL is a Datalog; also an HTTP Documents API
Schema on writeyes · Closed-world schema checked on every commit by default; schemaless raw_json mode is opt-in
ACID transactionsyes · Each commit atomic with snapshot isolation; a request can carry many document or WOQL writes
Storagelocal disk · Immutable succinct delta layers on the local storage path
Scale-outpartial · Single node; HA option over shared NFS storage; clustering is Enterprise only
Vector searchpartial · VectorLink 2.0 ANN search ships only in the 12.1 release candidate
Managed cloudpartial · Managed TerminusDB Cloud by Gettier is early access with design partners
+ All 24 sourced facts
LicenseApache-2.0 · Enterprise edition is commercial
Written inProlog, Rust · Engine core is SWI-Prolog; storage layer and 12.1 web server in Rust
First release2019 · Version 1.0 released October 2019
Schema migrationsyes · Migration API renames fields, casts types, adds defaults and transforms instance data
Concurrent writersyes · Optimistic concurrency, no locks; conflicting commits on a branch retry
Branch and mergeyes · Branches with field-level three-way merge (apply), conflict reports, rebase, squash, reset
Versioned readsyes · Query any earlier commit directly
Object storageno · Succinct delta layers on the local storage path
Larger than RAMno · Data persists on disk but queries run in memory over succinct layers; docs call it an in-memory database
Full-text searchpartial · Full-text search arrives with VectorLink in the 12.1 release candidate
Hybrid searchpartial · 12.1 RC search endpoint fuses vector and full-text; separate from WOQL graph queries
Embeddedno · Server only; the CLI is not an embeddable library
Serveryes · Docker, Snap, Kubernetes or source build
MCP serverpartial · Local MCP server added in the 12.1 release candidate
Agent frameworksno · No LangChain, LlamaIndex or similar integration found in docs
Runs without the vendoryes · Apache-2.0 self-host, no licence key

Neo4j Community Edition

GPL-3.0 · Java, Scala · 17.4k stars · Run log (2026.10.2)

What it is
The property-graph database that introduced Cypher, written in Java and Scala. Version 2026.10 uses calendar versioning with Cypher 25, which Neo4j says supports most mandatory GQL features.
Best for
Cypher with the largest set of drivers, tools and tutorials. Vector and Lucene full-text indexes are in Community Edition, there’s an official MCP server, and LangChain, LlamaIndex and others integrate with it.
Watch out for
Community Edition is single-instance: clustering, online backup, and existence, type and key constraints are Enterprise-only. In the run, CE accepted both bad writes and refused to create a property type constraint.
Data modelproperty graph
Query languagesCypher, GQL · Cypher 25 supports the majority of mandatory GQL features, per Neo4j
Schema on writepartial · CE has uniqueness constraints only; existence, type, key constraints and graph types are Enterprise only
ACID transactionsyes · ACID transactions in CE
Storagelocal disk · Aligned store format in CE; block format Enterprise only
Scale-outno · CE is single-instance; clustering Enterprise only, sharding Infinigraph only
Vector searchyes · Vector indexes in CE; SEARCH clause since 2026.01
Managed cloudyes · Neo4j AuraDB
+ All 24 sourced facts
LicenseGPL-3.0 · Enterprise Edition is commercial, source closed
Written inJava, Scala · Per GitHub language stats
First release2007 · Started as open-source project in 2007; 1.0 shipped 2010
Schema migrationspartial · DDL for constraints and indexes; Neo4j-Migrations is a Labs project, not core
Concurrent writersyes · Lock-based concurrent transactions, read-committed isolation
Branch and mergeno · No branching in the CE or EE feature list
Versioned readspartial · CE has offline dump/load only; online backup is Enterprise only
Object storageno · Local store files; object storage only for backups
Larger than RAMyes · Disk-native store with a page cache
Full-text searchyes · Lucene full-text indexes in CE
Hybrid searchpartial · Traversal, vector and full-text in one Cypher query but no built-in rank fusion; fusion in neo4j-graphrag
Embeddedyes · Embeddable in a JVM process via the Java API
Serveryes
MCP serveryes · Official neo4j/mcp server (Go)
Agent frameworksyes · LangChain, LlamaIndex and others listed by Neo4j
Runs without the vendoryes · CE needs no licence key
What it is
A distributed graph database in Go with DQL and a native GraphQL layer, sharded by predicate over the Badger key-value store. Since v25, formerly enterprise features (CDC, ACLs, binary backups) ship in the open-source build.
Best for
Scale-out with distributed ACID transactions (snapshot isolation, Jepsen-tested) and a GraphQL API without a separate server. Vector search with HNSW and an MCP endpoint are built in.
Watch out for
By default, unknown predicates are created on write: the run accepted the undeclared nickname until Dgraph ran in strict mode. No managed cloud (the current owner, Istari Digital, says none is planned) and no embedded mode.
Data modelproperty graph · Triples (RDF N-Quad or JSON input) with facets on edges; no SPARQL
Query languagesDQL, GraphQL
Schema on writepartial · Default mode auto-creates unknown predicates; opt-in strict mode rejects them
ACID transactionsyes · Distributed ACID with snapshot isolation, Jepsen tested
Storagelocal disk · Badger LSM key-value store
Scale-outyes · Predicates sharded into tablets across Alpha groups, Zero rebalances
Vector searchyes · float32vector predicates with HNSW index, similar_to
Managed cloudno · Hypermode shut Dgraph Cloud; Istari says no cloud service planned
+ All 24 sourced facts
LicenseApache-2.0 · v25 removed the proprietary enterprise code and licence
Written inGo
First release2016 · Repo created 2015; first GitHub release v0.3, May 2016
Schema migrationspartial · Schema alter with background reindex; no migration tooling
Concurrent writersyes · Concurrent transactions; conflicts abort at commit
Branch and mergeno · None found
Versioned readspartial · Binary backups restored, not queried
Object storageno · Badger LSM store on local disk
Larger than RAMyes · Disk-backed Badger LSM store
Full-text searchyes · fulltext index with stemming and stop words; no BM25 ranking
Hybrid searchpartial · Vector, text filters and traversal in one DQL query, no ranking fusion
Embeddedno · modusGraph embedded project archived
Serveryes · Zero and Alpha processes
MCP serveryes · Built into Alpha at /mcp and /mcp-ro, or dgraph mcp
Agent frameworksno · No official framework integrations found
Runs without the vendoryes · Self-hosted only, no licence key since v25

NebulaGraph

Apache-2.0 · C++ · 12.5k stars · Run log (3.8.0)

What it is
A distributed C++ property-graph database with separate graph, meta and storage services over RocksDB and Raft, queried with nGQL and an openCypher subset.
Best for
Large graphs hash-sharded across storage nodes, with a strongly typed schema enforced by default. Both bad writes were rejected.
Watch out for
The docs state that nGQL doesn’t support transactions. The latest open-source release is 3.8.0 from May 2024; new features (GQL, vector search) go to the closed Enterprise 5.x line. Full-text search needs an external Elasticsearch cluster, and it’s three services to run, not one.
Data modelproperty graph · Directed property graph of tags and edge types
Query languagesnGQL, openCypher · openCypher 9 DQL subset inside nGQL; GQL only in Enterprise 5.x
Schema on writeyes · Strong-typed schema; tags and edge types must exist before insert
ACID transactionsno · Docs: nGQL does not support transactions
Storagelocal disk · RocksDB key-value per partition with Raft replication
Scale-outyes · Hash-sharded by vertex ID into partitions across storage nodes
Vector searchno · Vector search is Enterprise 5.x only
Managed cloudyes · NebulaGraph Cloud, pay as you go, 14-day trial; runs Enterprise 5.x features
+ All 24 sourced facts
LicenseApache-2.0 · Enterprise 5.x is commercial
Written inC++
First release2019 · v0.1.0 on GitHub, May 2019; 1.0 GA 2020
Schema migrationspartial · ALTER TAG and ALTER EDGE only, no migration tooling
Concurrent writerspartial · Parallel writers with no transactional isolation; each partition's Raft leader orders writes
Branch and mergeno · None found
Versioned readspartial · Snapshots and BR backups restored, not queried
Object storageno · RocksDB per partition on local disk
Larger than RAMyes · Disk-based RocksDB storage
Full-text searchpartial · Needs an external Elasticsearch cluster plus listener
Hybrid searchno · No vector search in Community; hybrid retrieval is a Cloud and Enterprise feature
Embeddedno · NebulaGraph Lite starts a single-node server, not in-process
Serveryes · graphd, metad, storaged services
MCP serveryes · Announced on vendor blog, repo in nebula-contrib; 3.x only, last push Mar 2025
Agent frameworkspartial · LangChain and LlamaIndex graph stores maintained in those projects, not by vesoft
Runs without the vendoryes · Community self-hosts with no licence key
What it is
A Rust graph database (Apache-2.0 since May 2026, AGPL-3.0 before) with vector and BM25 search on SlateDB. It runs on S3-compatible storage, local disk or in memory, as a server or embedded in Rust, TypeScript or Python.
Best for
A graph plus vector and keyword search on object storage, with traversal, ANN and BM25 in one transaction. Besides its own query DSL, the server answers openCypher at /v2/cypher; in the run it returned the same three projects.
Watch out for
Version 3 dropped declared schemas, so any label or property is accepted, and the run accepted both bad writes. No built-in rank fusion. The official MCP endpoint and auto-scaling readers are Helix Cloud only; the self-hosted server is a single node.
Data modelproperty graph · Labeled property graph, one label per node or edge
Query languagesHelix query DSL (own), openCypher · Native operation-tree DSL via SDKs; openCypher subset needs server image v0.0.13+ or source build
Schema on writepartial · No declared types; any label or property accepted; optional unique node indexes only
ACID transactionsyes · Each request is one serializable transaction; no BEGIN/COMMIT across requests
Storageobject storage, local disk · SlateDB on S3-compatible store or local directory; dev default is in-memory
Scale-outpartial · Self-hosted server is single node; embedded read-only handles share object storage; auto-scaling readers are Cloud only
Vector searchyes · ANN on node and edge properties, prefiltered by traversal
Managed cloudyes · Helix Cloud
+ All 24 sourced facts
LicenseApache-2.0 · AGPL-3.0 until May 2026
Written inRust
First release2025 · v1.0.0 Feb 2025; repo opened Nov 2024
Schema migrationsno · No schema to evolve; only index create and drop with backfill
Concurrent writerspartial · Single writer process serialises commits; conflicts fail at commit
Branch and mergeno · None in docs
Versioned readsno · Backups and point-in-time recovery on roadmap
Object storageyes · SlateDB on S3-compatible storage is the source of truth; memory and NVMe are caches
Larger than RAMyes · Object storage is source of truth; memory and NVMe are caches
Full-text searchyes · BM25 text indexes on node and edge properties
Hybrid searchpartial · Traversal, vector and BM25 in one transaction; no built-in rank fusion
Embeddedyes · Rust, TypeScript, Python embedded runtimes
Serveryes · Standalone Docker server
MCP serverno · Official MCP endpoint is Helix Cloud only; CLI adds a docs MCP
Agent frameworksno · None found in docs
Runs without the vendoryes · Apache-2.0, no licence key

JanusGraph

Apache-2.0 · Java · 5.9k stars · Run log (1.2.0)

What it is
A distributed Java property-graph database under the Linux Foundation, queried with Gremlin. It’s a layer over a storage backend you choose (Cassandra, ScyllaDB, HBase, Bigtable, BerkeleyDB JE or in-memory), plus an optional index backend for full-text search.
Best for
Teams that already run Cassandra, ScyllaDB or HBase and want a graph on top that scales out with them. No company, paid edition or license key.
Watch out for
The docs say transactions aren’t necessarily ACID: they can be on BerkeleyDB, and generally aren’t on Cassandra or HBase. No vector search in 1.2.0. By default it creates property keys on first use, so the run accepted nickname until the strict settings were on. The run used BerkeleyDB JE with Lucene.
Data modelproperty graph · Stored as adjacency lists in a Bigtable-model backend
Query languagesGremlin · Native TinkerPop Gremlin; Cypher or SPARQL only via third-party translators to Gremlin
Schema on writepartial · Default creates labels and property keys on first use; with schema.default=none and schema.constraints=true undeclared types and wrong data types fail
ACID transactionspartial · Docs: transactions are not necessarily ACID; can be on BerkeleyDB, generally not on Cassandra or HBase
Storagebackend-dependent · Cassandra, ScyllaDB, HBase, Bigtable, BerkeleyDB JE or in-memory
Scale-outyes · Sharding comes from the backend cluster (Cassandra, HBase, Bigtable); stateless JanusGraph instances scale out
Vector searchno · No vector index or kNN predicate in 1.2.0; proposal moved to a GitHub discussion
Managed cloudno · None from the project; docs cover running on managed Cassandra (Astra DB, Amazon Keyspaces experimental)
+ All 24 sourced facts
LicenseApache-2.0 · Code Apache-2.0; docs CC-BY-4.0 (Apache-2.0 for docs imported from Titan)
Written inJava · Per GitHub language stats
First release2017 · v0.1.0 on GitHub, April 2017; forked from Titan on 27 Dec 2016
Schema migrationspartial · Schema elements can be added or renamed, not redefined; new indexes need a reindex job (local or MapReduce); no migration tooling
Concurrent writersyes · Many JanusGraph instances write to one backend cluster; optional per-element LOCK consistency; isolation backend-dependent
Branch and mergeno · None found
Versioned readsno · No versioned reads in the docs; backups are left to the storage backend
Object storageno · Rows live in the storage backend (Bigtable data model); no object-storage store
Larger than RAMyes · Disk-backed backends; BerkeleyDB needs only hot data in memory; in-memory backend is limited to the JVM heap
Full-text searchpartial · Needs an index backend: Elasticsearch/OpenSearch or Solr (external) or Lucene (local)
Hybrid searchpartial · Traversal plus full-text predicates (textContains etc.) in one Gremlin query; no vector search, no rank fusion
Embeddedyes · Embeds as a library in a JVM application (janusgraph-core plus backend modules)
Serveryes · JanusGraph Server, built on TinkerPop Gremlin Server
MCP serverpartial · No JanusGraph MCP server; generic Gremlin MCP servers only (kpritam/gremlin-mcp lists JanusGraph as untested; Apache TinkerPop gremlin-mcp is on the 4.0 pre-release line)
Agent frameworkspartial · LangChain community GremlinGraph, generic Gremlin tuned for Azure Cosmos DB; nothing from the project
Runs without the vendoryes · Self-hosted, no license key or commercial license
What it is
A PostgreSQL extension that adds openCypher over a property graph stored in ordinary PostgreSQL tables. Cypher runs inside SQL through the cypher() function.
Best for
Adding graph queries to a PostgreSQL you already operate. Transactions, MVCC, backups and replication are PostgreSQL’s, and SQL and Cypher mix in one statement. Vector search comes from pgvector, whose functions work inside Cypher.
Watch out for
Properties are a free JSON-like map with no type checks, so both bad writes went in. No sharding: the AGE FAQ says its tables can’t be distributed with Citus yet. Full-text search is SQL-side, not in Cypher.
Data modelproperty graph · Labeled property graph next to relational tables in the same database
Query languagesopenCypher, SQL · Cypher is wrapped in SQL via cypher(); SQL and Cypher mix in one statement
Schema on writeno · Labels are created on first CREATE; properties are a free JSON-like map with no type checks
ACID transactionsyes · From PostgreSQL; FAQ states ACID for graph, relational and JSON data
Storagelocal disk · PostgreSQL tables in the graph's schema
Scale-outpartial · No sharding: FAQ says AGE tables can't be distributed with Citus yet; PostgreSQL read replicas work (1.8.0 fixes VLE on read-only replicas)
Vector searchpartial · Via the separate pgvector extension; its distance functions and operators work inside Cypher; use of HNSW indexes from Cypher not documented
Managed cloudno · No ASF-run service; third parties host it, e.g. Azure Database for PostgreSQL offers AGE as an extension
+ All 24 sourced facts
LicenseApache-2.0
Written inC · Per GitHub language stats; parser in Yacc/Lex
First release2020 · v0.1.0 tagged 2020; entered the Apache Incubator 2020-04-30; based on Bitnine's AgensGraph
Schema migrationsno · No property schema to evolve; only create/drop of graphs and labels
Concurrent writersyes · From PostgreSQL MVCC; label tables are regular heap tables
Branch and mergeno · None in AGE or PostgreSQL docs
Versioned readspartial · No versioned reads; PostgreSQL point-in-time recovery restores to a timestamp
Object storageno · PostgreSQL data directory on local disk
Larger than RAMyes · Disk-based PostgreSQL tables; graph size limited by ID space, not RAM
Full-text searchpartial · Not in AGE; PostgreSQL core tsvector search is SQL-side; only pg_trgm similarity is tested inside Cypher
Hybrid searchpartial · MATCH plus pgvector ordering in one Cypher query; no text-vector rank fusion
Embeddedno · Requires a PostgreSQL server
Serveryes · Runs inside a PostgreSQL server you operate
MCP serverpartial · Community servers only, e.g. rioriost/age_mcp_server
Agent frameworkspartial · LangChain AGEGraph in langchain-community; no ASF-maintained integration
Runs without the vendoryes · Apache-2.0, no licence key or vendor
What it is
A strongly typed database with its own type-theoretic model (entities, n-ary relations and attributes with inheritance) and the TypeQL language. Rewritten in Rust for 3.0, on RocksDB.
Best for
Domains where the model is the hard part: relations between more than two things, type hierarchies, and every write validated against the schema.
Watch out for
Community Edition is single-server; replication is Cloud and Enterprise only. No full-text index, and native vector search is an alpha for invited testers. No LangChain or LlamaIndex package.
Data modeltype-theoretic · Entities, n-ary relations and attributes with type inheritance
Query languagesTypeQL
Schema on writeyes · Every write is validated against the defined schema
ACID transactionsyes · ACID up to snapshot isolation; read, write and schema transactions
Storagelocal disk, managed service · RocksDB on local disk; TypeDB Cloud managed
Scale-outno · CE single-server only; Cloud/Enterprise replicate full copies via Raft, no sharding
Vector searchno · Native vector type is an alpha for invited testers (Aug 2026)
Managed cloudyes · TypeDB Cloud on AWS and Google Cloud
+ All 24 sourced facts
LicenseMPL-2.0 · Cloud and Enterprise are commercial
Written inRust · 3.0 rewrite; 2.x was Java
First release2021 · TypeDB 2.0 in April 2021, renamed from Grakn (first release 2016)
Schema migrationspartial · define, undefine, redefine in schema transactions; version moves go through export and import
Concurrent writersyes · Concurrent write transactions, conflicting commit fails; schema transactions take an exclusive lock
Branch and mergeno · No branches of the data
Versioned readspartial · No versioned reads; backups are DIY via export and import in CE
Object storageno · RocksDB on local disk
Larger than RAMyes · RocksDB-backed; 3.14.1 notes recommend RAM around 10% of on-disk size
Full-text searchno · String contains and like comparisons only, no text index
Hybrid searchno · No vector or full-text index to combine
Embeddedno · Server accessed by drivers over gRPC or HTTP
Serveryes
MCP serveryes · Official typedb/typedb-mcp, Apache-2.0, Docker image
Agent frameworksno · No LangChain, LlamaIndex or other framework package found
Runs without the vendoryes · CE self-hosts with no key; only clustering needs Cloud or Enterprise

LadybugDB

MIT · C++ · 1.9k stars · Run log (0.22.3)

What it is
The community fork of Kùzu, started in October 2025 after Kùzu’s repo was archived. An embedded, columnar, single-file graph database with Cypher; it also runs in Wasm.
Best for
Analytics and local graphs inside your application, with no server to run. Graphs are strictly typed by default, and both bad writes were rejected.
Watch out for
One write transaction at a time, and one read-write process per database file. No scale-out. Vector and full-text search are official extensions you INSTALL and LOAD.
Data modelproperty graph · Structured property graph: node and rel tables
Query languagesCypher
Schema on writeyes · Graphs are strictly typed by default; CREATE GRAPH ... ANY opts into open types
ACID transactionsyes · Serializable; BEGIN TRANSACTION, COMMIT, ROLLBACK
Storagelocal disk, in-memory · Single .lbdb file; Icebug Parquet tables can live on S3
Scale-outno · Embedded, single process
Vector searchpartial · Disk HNSW, but only as the official vector extension you INSTALL
Managed cloudno · None found
+ All 24 sourced facts
LicenseMIT
Written inC++
First release2025 · Fork repo created Oct 2025; Kuzu repo dates from 2020
Schema migrationspartial · ALTER TABLE add, drop, rename columns; no migration tooling
Concurrent writerspartial · One write transaction at a time; one read-write process per database file
Branch and mergeno · None in docs
Versioned readsno · No versioned reads; EXPORT DATABASE dumps only
Object storagepartial · Database is one local .lbdb file; Icebug Parquet tables can live on S3
Larger than RAMyes · Disk-based columnar storage with buffer manager; spills to disk
Full-text searchpartial · BM25, but only as the official fts extension you INSTALL; node tables only
Hybrid searchpartial · Vector search plus Cypher traversal in one query; no text-vector fusion
Embeddedyes · In-process library, also Wasm
Serveryes · Official Bolt, ADBC/Arrow Flight and REST API servers
MCP serveryes · mcp-server-ladybug in the LadybugDB org
Agent frameworkspartial · LangChain and LlamaIndex packages are community-maintained; docs link notebooks
Runs without the vendoryes · MIT, no licence key
What it is
A multi-model database in Java (graph, document, key/value, time series, vector) from the founder of OrientDB. One edition, with no feature gating. It speaks SQL, openCypher, Gremlin, GraphQL and the MongoDB and Redis protocols, plus Bolt and the Postgres wire protocol.
Best for
One engine for several models and query languages, with ACID transactions, Raft replication in the open-source edition, vector and full-text indexes and a built-in MCP server.
Watch out for
Declared properties are validated, but undeclared ones are accepted, and we found no stricter mode in the docs: the run rejected age: "forty" and accepted nickname. Replicas hold full copies, with no sharding.
Data modelmulti-model · Graph, document, key/value, time series, vector, geospatial
Query languagesSQL, openCypher, Gremlin, GraphQL, MongoDB QL · SQL is OrientDB's dialect; Redis commands; Bolt and Postgres wire protocols
Schema on writepartial · Declared properties validated (type, NOT NULL, regex, min/max); undeclared properties accepted; Cypher CREATE makes missing types implicitly
ACID transactionsyes · ACID with WAL; statement-level savepoints since 26.11.1
Storagelocal disk · Page-based files with a page cache
Scale-outpartial · Raft leader-replica HA in the open-source edition; every replica holds a full copy, no sharding
Vector searchyes · LSM vector index on JVector (HNSW or Vamana/DiskANN graphs)
Managed cloudno · No hosted service on the pricing page or site
+ All 24 sourced facts
LicenseApache-2.0 · Single edition; GOVERNANCE.md commits to Apache 2.0 with no CLA
Written inJava · Per GitHub language stats; Java 21+
First release2021 · First GitHub release 21.9.1, September 2021; conceptual fork of OrientDB, reuses its SQL engine
Schema migrationspartial · ALTER TYPE and ALTER PROPERTY on populated databases; no migration tooling found
Concurrent writersyes · Optimistic MVCC, conflicts raise ConcurrentModificationException at commit; READ_COMMITTED default, REPEATABLE_READ available
Branch and mergeno · None found in docs
Versioned readspartial · Online full backups you restore, not query
Object storageno · Local files; backups cannot even be written to S3 directly
Larger than RAMyes · Disk-based pages; maxPageRAM caps pages kept in RAM (default 25% of heap)
Full-text searchyes · Lucene-based FULL_TEXT index, BM25 by default
Hybrid searchpartial · vector.fuse fuses dense, sparse and full-text results (RRF, DBSF, linear) in one SQL query; no doc example combines it with graph MATCH in the same query
Embeddedyes · Runs in-process in any JVM language; Python bindings
Serveryes · Docker, Kubernetes, binary distribution
MCP serveryes · Built-in MCP server module, in the standard distribution
Agent frameworkspartial · Official langchain-arcadedb only; Haystack and LangChain4j (via Bolt) are not on the list; no LlamaIndex found
Runs without the vendoryes · No licence key, node limits or feature gating

OrientDB

Apache-2.0 · Java · 5.0k stars · Run log (3.2.58)

What it is
A Java multi-model database (graph, document, key-value, object) queried with an extended SQL with MATCH patterns, and Gremlin through a TinkerPop package. After SAP ended its maintenance in 2023, it continues as a community project with roughly monthly 3.2.x releases.
Best for
Existing OrientDB users, and embedded or server multi-model use in Java with Lucene full-text search. The former Enterprise Agent (incremental backup, cluster management) is now free under Apache-2.0.
Watch out for
Classes are schema-less or hybrid by default; undeclared properties are rejected only in strict mode. No vector index. Sharding is manual and the docs flag its limitations. Version 4.0 is in development with no beta yet.
Data modelmulti-model · Graph, document, key-value and object in one engine
Query languagesSQL, Gremlin · OrientDB SQL with MATCH pattern queries; Gremlin via the TinkerPop 3 (Gremlin Server) package
Schema on writepartial · Schema-less by default; per-class strict mode rejects undefined properties, typed and mandatory properties optional (schema-full or hybrid per class)
ACID transactionsyes · ACID with MVCC; READ COMMITTED default, REPEATABLE READ on plocal and memory; tx held in client RAM
Storagelocal disk, in-memory · plocal paginated disk storage, or memory storage
Scale-outpartial · Multi-master replication; manual class-level sharding over clusters, app picks the shard, sharded indexes unsupported, docs flag limitations
Vector searchno · Index types are SB-Tree, hash, auto-sharding, Lucene full-text and spatial; no vector index in docs or repo issues
Managed cloudno · No hosted offering; site lists only downloads and support plans
+ All 24 sourced facts
LicenseApache-2.0 · Enterprise Agent also Apache-2.0 since SAP open-sourced it
Written inJava · Requires Java 17 on develop
First release2010 · Java rewrite of Orient ODBMS (C++, 1999); GitHub repo created Dec 2012
Schema migrationspartial · ALTER CLASS / ALTER PROPERTY DDL; schema changes are not transactional; no migration tooling
Concurrent writersyes · Optimistic MVCC with record versions; conflicting commit fails, no global write lock
Branch and mergeno · No branches of the data
Versioned readspartial · Full backups plus incremental backups through the now-free Enterprise Agent; restore, not query
Object storageno · plocal files on local disk
Larger than RAMyes · plocal is page-based on disk with a 2Q read cache and write cache in front
Full-text searchyes · Lucene full-text index with SEARCH_CLASS / SEARCH_INDEX functions
Hybrid searchpartial · SQL can mix traversal and Lucene full-text in one query; no vector search, no fusion
Embeddedyes · Open databases in-process with the embedded: URL, or embed the full OServer in a Java app
Serveryes · Standalone server with binary and HTTP protocols, Studio web UI
MCP serverno · No official or community OrientDB MCP server found on GitHub
Agent frameworksno · No LangChain or LlamaIndex integration found on GitHub
Runs without the vendoryes · Apache-2.0 builds and free agent.jar from Maven Central, no licence key; SAP support already ended

[Licenses]

Often called open source: six source-available engines

You can read their code, and most can be self-hosted for free, but their licenses aren’t OSI-approved and restrict some uses, usually offering the database as a service.

EngineLicenseWhat it means in practice
MemgraphBSL-1.1Converts to Apache-2.0 after four years. Community needs no key; RBAC, multi-tenancy and automatic failover need an Enterprise license.
FalkorDBSSPL-1.0Self-host the Redis module with no key; commercial license on request.
ArangoDBBSL-1.1Source-available since 3.12: internal production allowed, no competing commercial offering. The free prebuilt binaries use a separate Community License: non-commercial, read-only past 100 GiB.
SurrealDBBSL-1.1Converts to Apache-2.0 on 1 January 2030; no third-party database-as-a-service. SDKs are Apache-2.0.
FlureeBSL-1.1Free in production except as a hosted database service; each release converts to Apache-2.0 after four years. Has branching, merge and time travel.
HydraDBElastic-2.0AGPL-3.0 until 3 November 2026. The license bars offering it as a managed service; vector search, BM25 and the memory API are Cloud only.

Watched, not listed: Kùzu (Archived in Oct 2025; LadybugDB is the maintained fork. Mentioned on the page, not listed.) Cayley (Apache-2.0 but last release 2019. Fails the activity rule; listed again if it ships.) Not covered: managed and proprietary engines such as Amazon Neptune, TigerGraph and Google Spanner Graph.

[FAQ]

Questions

What is an open source graph database?

A graph database stores data as nodes and the relationships between them, and answers queries by following those relationships, in languages such as Cypher, Gremlin or GQL. It’s open source when its engine is released under a license the Open Source Initiative has approved, such as Apache-2.0, MIT, GPL-3.0 or MPL-2.0. Licenses like BSL, SSPL and the Elastic License let you read the code but restrict use, so they’re source-available. Check the LICENSE file of the engine itself, not just the SDKs: several vendors use permissive licenses for clients and a restrictive one for the server.

Is Neo4j open source?

Neo4j Community Edition is, under GPL-3.0. Enterprise Edition is commercial and closed source, and several features are Enterprise-only, including clustering, online backup and property type constraints. In our run, Community Edition refused to create a type constraint.

What’s the best open source graph database for AI agents?

Pick by what the agents do with the graph. If several agents write and their changes should be reviewed or rolled back, you want branches and versioned reads: Omnigraph and TerminusDB are the open-source engines with both. For the widest framework support, Neo4j Community Edition has LangChain and LlamaIndex integrations and an official MCP server. For a graph inside your own process, LadybugDB.

Can I run a graph database on S3?

Yes. Among open-source engines, Omnigraph stores the graph as Lance datasets on S3-compatible storage, GCS or R2, and HelixDB uses SlateDB on S3-compatible storage. LadybugDB can keep Parquet tables on S3, but the database itself is a local file. Fluree and HydraDB also run on object storage, under source-available licenses.

Which open source graph databases can run embedded?

Seven of the twelve run inside your process. LadybugDB is an in-process library with a single-file database, and also runs in Wasm. Omnigraph ships a Rust engine crate, and its CLI runs in-process against a local graph. HelixDB has embedded runtimes for Rust, TypeScript and Python. Neo4j Community Edition, JanusGraph, ArcadeDB and OrientDB embed in a JVM process (ArcadeDB also has Python bindings). Dgraph, NebulaGraph, TypeDB, TerminusDB and Apache AGE need a server.

What replaced Kùzu?

Kùzu’s repository was archived in October 2025. LadybugDB is the maintained community fork, under MIT, with the same embedded, columnar design and Cypher.

[Changelog]

What changed

First version. 12 open-source engines, 6 source-available. HydraDB listed as source-available after its 3 November relicense from AGPL-3.0 to Elastic License 2.0.