Skip to content
Kuzu
Blog

Differences between Kuzu and Neo4j

If you’re coming over from Neo4j, you can find the differences from Neo4j’s keywords and syntax in this section. Note that as far as possible, Kuzu tries to follow openCypher’s syntax and semantics.

Schema

Unlike Neo4j, Kuzu requires a schema to be defined before any data can be inserted into the database (we call this the “structured property graph model”). The schema provides a logical grouping of node and relationship tables, along with their associated properties and data types that define the structure of the graph database. See our DDL docs for more information.

Clauses

CREATE and MERGE

It is recommended to always specifying node and relationship labels explicitly in the CREATE and MERGE clause. If not specified, Kuzu will try to infer the label by looking at the schema.

FINISH

FINISH is recently introduced in GQL and adopted by Neo4j but not yet supported in Kuzu. You can use RETURN COUNT(*) instead which will only return one record.

FOREACH

FOREACH is not supported. You can use UNWIND instead.

LOAD CSV FROM

Kuzu can scan files not only in the format of CSV, so the LOAD CSV FROM clause is renamed to LOAD FROM.

MATCH and OPTIONAL MATCH

Relationship cannot be omitted. For example --, --> and <-- are not supported. You need to use -[]-, -[]-> and <-[]- instead.

Semantics

Neo4j adopts a trail semantic (no repeated edge) for pattern within a MATCH clause, whereas Kuzu adopts walk semantic (allowing repeated edge) for pattern within a MATCH clause.

In Kuzu, you can use is_trail or is_acyclic function to check if a path is a trail or acyclic.

Variable length relationships

Since Kuzu adopts trail semantic by default, so a variable length relationship needs to have a upper bound to guarantee the query will terminate. If upper bound is not specified, Kuzu will assign a default value of 30.

Kuzu also extends Neo4j’s variable length to support filter inside the variable length relationship.

To run algorithms like (all) shortest path, simply add SHORTEST or ALL SHORTEST between the kleene star and lower bound. For example, MATCH (n)-[r* SHORTEST 1..10]->(m). It is recommended to use SHORTEST if paths are not needed in the use case.

More information can be found in the MATCH documentation page.

REMOVE

REMOVE is not supported. Use SET n.prop = NULL instead.

SET

Properties must be updated in the form of n.prop = expression. Update all properties with map of += operator is not supported. Try to update properties one by one.

USE

USE graph is not supported. For Kuzu, each graph is a database. You can use different graph by opening different databases.

WHERE

Using WHERE inside node or relationship pattern is not supported, e.g. MATCH (n:Person WHERE a.name = 'Andy') RETURN n. You need to write it as MATCH (n:Person) WHERE n.name = 'Andy' RETURN n.

Filter on node or relationship labels is not supported, e.g. MATCH (n) WHERE n:Person RETURN n. You need to write it as MATCH (n:Person) RETURN n, or MATCH (n) WHERE label(n) = 'Person' RETURN n.

Others

Any SHOW XXX clauses become a function call in Kuzu. For example, SHOW FUNCTIONS in Neo4j is equivalent to CALL show_functions() RETURN * in Kuzu. For more information, see the functions page.

Subqueries

Kuzu supports EXISTS and COUNT subquery. See Subqueries for more information.

CALL <subquery> is not supported.

Data Types

Kuzu following Postgres typing system. For LIST type, all elements should be of the same type. For MAP type, all keys should be of the same type and all values should be of the same type. For more information, see data types.

Functions

  • labels() function is named as label() instead.
  • type predicate expression n.property IS :: INTEGER is supported as typeOf(n.property)=INT64
  • internal id function elementId is supported as id().
  • spatial functions are not supported.

Aggregate functions

  • percentileCont, percentileDisc, stDev and stDevP are supported.

List functions

  • Most list functions have a list_ prefix, e.g. list_concat, list_reverse, list_reduce, etc…
  • tail is supported as list_slice().
  • head and tail are supported as list_extract() or list[].

Casting functions

  • toXXX functions in Neo4j are supported in Kuzu via the cast(input, targetType) function

Mathematical functions

  • isNaN is not supported.
  • e() is not supported.
  • haversin is not supported.
  • pi() is not supported.

String functions

  • char_length and character_length are supported as size.

Temporal functions

  • date() is supported as current_date(). timestamp() is supported as current_timestamp().
  • local datetime, real time clock, transaction time clock are not supported.

Vector similarity functions

  • cosine similarity is named as ARRAY_COSINE_SIMILARITY()
  • euclidean distance is named as ARRAY_DISTANCE()

Indexes and Constraints

Kuzu does not currently support manually creating indexes or constraints on custom properties. Instead, Kuzu creates a primary key index (which also guarantees non-null and uniqueness) the user-specified primary key column of a node table.