Raw Cypher

⌘4. The bar turns mono and runs exactly what you type, read-only. No AI is involved; the query is saved in history as its own question.

Writing it

The field highlights as you type: keywords, labels, strings and numbers in their own colours; a label the graph doesn’t have is red with a dotted underline. Pasted text with line breaks is put on one line.

Completion opens as you type and shows, by what the cursor is in:

• after a : or | inside ( ): Node types; inside [ ]: Relationship types; elsewhere, both;
• after variable.: Properties of n (narrowed to the labels you gave n in the pattern: (v:Verse) offers Verse’s columns, each with its shape name and kind: “Translation · Text”; id is always there);
• a word anywhere else: the query’s own variables (from patterns and AS), keywords (MATCH, OPTIONAL MATCH, WHERE, RETURN, WITH, ORDER BY, LIMIT, SKIP, DISTINCT, CONTAINS, STARTS WITH, IS NOT NULL, UNWIND, CASE…), and functions with what they do (count how many, collect into a list, label a node’s or relationship’s type, length a path’s length, nodes, rels, coalesce…).

Tab completes the first (or highlighted) row; Return still runs the query unless you’ve moved to a row with the arrows. A function comes with both brackets and the cursor between them. Nothing is offered inside a string or a comment, and a number isn’t completed.

Names to use

The database names from your shape: FreeEnergyPrinciple, RELATES_TO, year_published, CONTAINS_ (shown as CONTAINS in the app). The completion lists them; the Cypher block in a Text-to-Cypher trail shows them in use; the MCP get_shape tool prints them all.

The guard

Before anything runs, a keyword check refuses CREATE, MERGE, SET, DELETE, DETACH, REMOVE, DROP, ALTER, COPY, INSTALL, LOAD, ATTACH, USE, EXPORT, IMPORT, BEGIN, COMMIT, ROLLBACK, CHECKPOINT and CALL, and more than one statement: Not run: Ask only reads your graph. This query uses DELETE, which would change it. Then the query runs inside a read-only transaction, the second line of defence. At most 500 rows are kept (“Stopped at 500 rows. The query matches 2,314.”); the trail reports the count and the time.

Keywords are only refused where they’re keywords: a property called set (n.set) or a label named Create is fine.

The dialect

LadybugDB speaks the Kuzu dialect of Cypher. The differences from Neo4j that most examples are written for:

• label(n) gives a node’s or relationship’s label; there is no labels() or type().
• Variable-length paths are written -[:REL*1..3]-> and always need an upper bound; shortest paths are -[:REL* SHORTEST 1..6]->.
• A property missing on a node is NULL; test with IS NULL / IS NOT NULL.
• String functions include lower, upper, contains, starts_with, ends_with and regexp_matches.
• Aggregates group implicitly by the other returned columns.
• Every node and relationship has an id column, a UUID; match one exactly with MATCH (n {id: upper('…')}).

Results

Return whole nodes and relationships (RETURN c, r, s) and they light up on the canvas and list as a table; return scalars and you get a table of values, or a single big number for one value. See Answers on the canvas.