Repository navigation
openCypher: accept Neo4j's CREATE VECTOR INDEX and CREATE FULLTEXT INDEX, so Neo4j clients can create what db.index.* already queries #10678
Description
We write to ArcadeDB through cocoindex's stock Neo4j connector over Bolt. cocoindex has no ArcadeDB target, and does not plan one (cocoindex-io/cocoindex#603). We measured this on 26.10.1 and 26.10.2.
- Rows work:
MERGE … SET n += $props,DETACH DELETE, edgeMERGE, andCALL { … } IN TRANSACTIONS. - fix(#9646): DROP CONSTRAINT <name> undoes what CREATE CONSTRAINT <name> created; unknown name raises #9707 helped. 26.10.2 made the connector's
DROP CONSTRAINTwork. Thank you.
What fails is the schema. Every statement below is refused on both releases:
Statement | Answer -- | -- CREATE VECTOR INDEX v IF NOT EXISTS FOR (n:`L`) ON n.`e` OPTIONS { indexConfig: { `vector.dimensions`: 768, `vector.similarity_function`: 'cosine' } } | SyntaxError: Only standard, RANGE and TEXT index types are supported CREATE FULLTEXT INDEX f IF NOT EXISTS FOR (n:`L`) ON EACH [n.`text`] | same CREATE CONSTRAINT t FOR (n:`L`) REQUIRE n.`e` IS :: VECTOR(768) | Unsupported Cypher type in a property type constraint SQL CREATE INDEX ON L (e) LSM_VECTOR … after a MERGE wrote e as a list of floats | SchemaException: the property does not existSo today the client creates the typed ARRAY_OF_FLOATS property, LSM_VECTOR and FULL_TEXT in SQL over HTTP, before its first Bolt write.
The grammar already parses createVectorIndex and createFulltextIndex. CypherASTBuilder.handleCreateIndex is where they are refused. The read side is already Neo4j compatible through db.index.vector.queryNodes and db.index.fulltext.queryNodes; only the DDL that creates what they read is missing.
Proposal:
CREATE VECTOR INDEX. Declare the propertyARRAY_OF_FLOATSwhen it is undeclared, asCREATE INDEXalready declares or infers types, and buildLSM_VECTORfromvector.dimensionsandvector.similarity_function(cosine or euclidean). Keep the given name, as CypherCREATE INDEX <name>ignores the name, soDROP INDEX <name>is a silent no-op and an unknown name reports success #9705 does forCREATE INDEX, soDROP INDEX <name> IF EXISTSremoves it.CREATE FULLTEXT INDEX. BuildFULL_TEXT, taking the analyzer fromfulltext.analyzer.- Optionally,
IS :: VECTOR<FLOAT32>(n). Map it toARRAY_OF_FLOATS.
With these in place, a Neo4j client that creates its own vector index with Neo4j 5 DDL would run on ArcadeDB unchanged.
Activity
docs(arcadedb): record 26.10.2 in CI and on the lab VM
feat(#10678): openCypher accepts Neo4j's CREATE VECTOR INDEX and CREA…
robfrank commentedon Oct 28, 2026
Done in #10864 for the common forms. openCypher now accepts Neo4j 5's
CREATE VECTOR INDEX(becomes anLSM_VECTORindex; an undeclared property is declaredARRAY_OF_FLOATS, andvector.dimensions,vector.similarity_function, HNSW and quantization options are mapped) andCREATE FULLTEXT INDEX ... ON EACH [...](becomes aFULL_TEXTindex, analyzer fromfulltext.analyzer). Both keep the given name, honourIF NOT EXISTS, andSHOW INDEXESlists acreateStatementthat runs as written. SQLCREATE INDEXafter a MERGE still needs a declared property, by design; the Cypher DDL declares it itself.Still open:
IS :: VECTOR<FLOAT32>(n)type constraint: openCypher: accept IS :: VECTOR<T>(n) in a property type constraint, enforcing the dimension #10862.(n:A|B)),WITH [...]filter properties, a vector index without a dimension, thestandardanalyzer, the emptyoptionscolumn of SHOW INDEXES, and Neo4j's quantization-on default: openCypher: multi-label vector/full-text indexes, WITH filter properties, dimension-less vector index, 'standard' analyzer #10863. Each is refused today with an explicit message and never built differently.❤️ Thanks for helping make ArcadeDB better! ArcadeDB is free and open source, sustained by its community. If it's useful to you or your company, please consider becoming a sponsor to keep development going.
feat(#10678): openCypher accepts Neo4j's CREATE VECTOR INDEX and CREA…