ArcadeDB

ArcadeDB is a Java engine. It runs embedded inside a Java process or as a server behind HTTP, Bolt, Postgres, and Redis protocols, and almost everyone runs the server. HumemAI wraps the same engine for Python through JPype, so it also runs inside a Python process; that package is the Python section further down. It answers SQL, in its own dialect, and Cypher, so the two industry-standard query languages and the skills and tools around them carry over; the other single-stack multi-model engines each have a language of their own, SurrealQL, AQL, and MongoDB's query language. Upstream also lists key-value and full-text search among its models; this page does not cover them.

Most databases that call themselves multi-model are several engines behind one API. ArcadeDB is not. Everything it stores sits on the same pages, goes through the same write-ahead log (WAL), and commits in the same transaction, so a write that touches a document, an edge, and a vector index is one ACID transaction (atomic, consistent, isolated, and durable) instead of three that have to be coordinated.

The indexes work the same way. LSM trees (log-structured merge trees), full-text, geo, hash, and both dense and sparse vector indexes all commit in the same transaction as the records they index. Replication follows from that: Raft ships page changes from the shared log, so every model replicates without replication code per model.

Vectors are the one exception. The vector records are transactional, logged, and replicated like everything else, but the nearest-neighbour graph used to search them is not: it is built in the background and can be rebuilt. The data is the source of truth and the search structure catches up to it. That is more than a standalone vector store offers, and less than a fully transactional index would be.

The cross-engine benchmarks are their own project, DBBench, so that they stay independent of any one engine. This page keeps to ArcadeDB itself.

The first results are published in November 2026: ArcadeDB 26.11.1 against a multi-model subset of engines, measured on the November instrument. Until then DBBench shows its method and its lanes, and no numbers.

Not every engine that DBBench measures is open source in the sense the Open Source Initiative (OSI) defines. ArcadeDB and the Python package are both Apache-2.0.

Two tests, each answered from the license text at the pinned version in September 2026. Any use, free: anyone may run the engine for any purpose, including selling a product or a service built on it, without paying the vendor; the SSPL (MongoDB), the Business Source License (SurrealDB and ArangoDB), and the Timescale License (the TimescaleDB image used here; its core alone is Apache-2.0) all reserve some uses. Your code stays yours: you may modify the engine and ship or serve a product on it without publishing your own code; the GPL (Neo4j Community) and the AGPL (Elasticsearch, whose core is offered under a choice of three licenses since 2024 and whose sparse-vector search is in that core) require sharing what you distribute or serve. Rows in bold pass both tests. Both tests are yes for permissive licenses (Apache-2.0, MIT, the PostgreSQL License) and the public domain, yes and no for copyleft, and no for source-available. The Open Source Initiative approves every license in the first two groups and none in the third. A license can change with a release.

Four of those engines are single-stack multi-model databases, one process serving documents, graph, and vectors: ArcadeDB, ArangoDB, MongoDB, and SurrealDB. Among them ArcadeDB is the only one that passes both tests above, Apache-2.0 since its first commit in 2021, and the only one whose engine and Python bindings ship together in monthly releases; MongoDB moved from the AGPL to the SSPL in 2018, ArangoDB from Apache-2.0 to the Business Source License in 2024, and SurrealDB has used the Business Source License since its public release. On staying that way, ArcadeDB's founder wrote in February 2026 that the project will never change its license; this page reports that as his published statement, not as a guarantee. HumemAI, which maintains the Python bindings (arcadedb and arcadedb-jvm), makes the same commitment for them: the engine and its bindings both stay Apache-2.0. PostgreSQL with pgvector and AGE reaches the same one-transaction result by loading extensions into one process, and it is open source under both tests; it is not counted among the four because three projects maintain its three models.