Graph Reference

Cinch graph databases use Ladybug (a fork of Kuzu), which speaks a Cypher dialect that is mostly compatible with Neo4j but has important differences. Standard Neo4j drivers work out of the box over Bolt.

Schema-first: the biggest difference

Neo4j lets you create nodes on the fly without declaring a schema. Cinch requires you to define node and relationship tables before inserting data. Every node table must have a PRIMARY KEY column.

-- Define node tables
CREATE NODE TABLE Person(name STRING PRIMARY KEY, age INT64)
CREATE NODE TABLE City(name STRING PRIMARY KEY, population INT64)

-- Define relationship tables (must specify FROM and TO)
CREATE REL TABLE KNOWS(FROM Person TO Person, since INT64)
CREATE REL TABLE LIVES_IN(FROM Person TO City)

-- Now you can insert data
CREATE (a:Person {name: 'Alice', age: 30})
CREATE (b:Person {name: 'Bob', age: 25})

-- SERIAL for auto-incrementing IDs
CREATE NODE TABLE Event(id SERIAL PRIMARY KEY, title STRING, date DATE)

✓

IF NOT EXISTS

Use CREATE NODE TABLE IF NOT EXISTS to make schema setup idempotent. Safe to run on every connection.

Differences from Neo4j

Feature	Neo4j	Cinch (Ladybug/Kuzu)
Schema	Schema-free. Create nodes on the fly.	Schema-first. CREATE NODE TABLE / CREATE REL TABLE before inserting.
Primary keys	Optional. Internal IDs used by default.	Required. Every node table needs a PRIMARY KEY column.
Multiple labels	Nodes can have multiple labels.	Nodes have exactly one label (their table name).
Relationship tables	Relationships created inline.	Must declare with CREATE REL TABLE(FROM X TO Y).
REMOVE clause	REMOVE n.prop	Not supported. Use SET n.prop = NULL instead.
WHERE in patterns	MATCH (n:Person WHERE n.name = 'Alice')	Not supported. Use MATCH (n:Person) WHERE n.name = 'Alice'
Label in WHERE	WHERE n:Person	Not supported. Use MATCH (n:Person) or WHERE label(n) = 'Person'
Variable-length paths	Trail semantics (no repeated edges)	Walk semantics by default (allows repeated edges). Use TRAIL modifier for Neo4j behavior.
Type system	Flexible lists and maps	Stricter: LIST elements must be same type, MAP keys must be same type.

Variable-length paths

By default, variable-length path queries use walk semantics, which allows repeated nodes and edges. This differs from Neo4j's default trail semantics. Use the TRAIL or ACYCLIC modifiers to change this:

-- Default: walk semantics (allows repeated edges)
MATCH (a:Person)-[e:KNOWS*1..4]->(b:Person) RETURN a.name, b.name

-- Trail: no repeated edges (Neo4j default behavior)
MATCH (a:Person)-[e:KNOWS* TRAIL 1..4]->(b:Person) RETURN a.name, b.name

-- Acyclic: no repeated nodes
MATCH (a:Person)-[e:KNOWS* ACYCLIC 1..4]->(b:Person) RETURN a.name, b.name

-- Shortest path
MATCH (a:Person)-[e:KNOWS* SHORTEST 1..10]->(b:Person) RETURN a.name, b.name

Supported Cypher

Schema: CREATE NODE TABLE, CREATE REL TABLE, DROP TABLE, ALTER TABLE
Node operations: CREATE, MERGE, MATCH, DELETE, DETACH DELETE, SET
Relationship operations: CREATE, MATCH, DELETE with typed and directional relationships
Pattern matching: variable-length paths, shortest path, path expressions
Clauses: WITH, UNWIND, OPTIONAL MATCH, UNION, CALL
Aggregations: COUNT, SUM, AVG, MIN, MAX, COLLECT
Filtering: WHERE, AND, OR, NOT, IN, STARTS WITH, CONTAINS, regular expressions
Ordering: ORDER BY, SKIP, LIMIT
Indexes: CREATE INDEX, DROP INDEX
Constraints: uniqueness constraints
Parameters: $paramName syntax
Transactions: BEGIN, COMMIT, ROLLBACK (via driver API)

Bolt protocol

Feature	Details
Bolt versions	4.4, 5.0 through 5.7
Auth	Basic auth (username: `neo4j`, password: your token)
Concurrent connections	256 per database
Transactions	Auto-commit and explicit (BEGIN/COMMIT/ROLLBACK)
Bookmarks	Supported for causal consistency

Not supported

APOC procedures
Graph Data Science (GDS) library. Use Cypher path-finding instead.
Multiple labels per node
REMOVE clause (use SET prop = NULL)
WHERE inside MATCH patterns
FINISH clause
USE graph (each database is a separate graph)

See Graph for connection examples.