ladybugdb

name

ladybugdb

description

Expert guide for LadybugDB — an embedded, in-process property graph database using openCypher. Use this skill whenever the user is working with LadybugDB, writing Cypher queries for LadybugDB, using the `lbug` CLI, importing `real_ladybug` in Python, using `@ladybugdb/core` in Node.js, or building any application with LadybugDB. Also triggers when the user asks about LadybugDB schema design, graph algorithms (PageRank, Louvain), HNSW vector search, full-text search, ATTACH to PostgreSQL/DuckDB/Delta Lake, LLM embeddings with CREATE_EMBEDDING, or bulk data import/export with COPY FROM/TO. Use even if the user just says 'ladybug graph db' or pastes a .lbug file path.

LadybugDB

LadybugDB is an embedded, in-process property graph database — no server process required. It uses the openCypher query language with a required, predefined schema (unlike Neo4j), columnar disk-based storage, vectorized query execution, and serializable ACID transactions.

Quick orientation

• Schema-first: you must create node/rel tables before inserting data
• One primary key per node table — automatically indexed, unique, non-null
• Walk semantics: repeated edges allowed in MATCH (unlike Neo4j's trail semantics)
• One write transaction at a time; multiple concurrent reads are fine
• In-memory mode: use ":memory:" as the database path for ephemeral databases

Installation

# CLI
curl -s https://install.ladybugdb.com | bash   # Linux
brew install ladybug                             # macOS

# Python
pip install real_ladybug

# Node.js
npm install @ladybugdb/core

CLI basics

lbug mydb.lbug        # open/create on-disk DB
lbug                   # in-memory (ephemeral)
lbug mydb.lbug < schema.cypher   # batch mode

Key shell commands: :schema (show tables), :help, :quit, :mode [json|csv|markdown|...]

Reference files

Load only the sections you need:

File	Contents
references/cypher-reference.md	DDL, DML, MATCH queries, transactions, macros, LadybugDB vs Neo4j differences
references/python.md	Python (real_ladybug) — connection, query, DataFrame, transactions
references/nodejs.md	Node.js (@ladybugdb/core) — connection, query, streaming, transactions
references/java.md	Java — Maven setup, connection, query, transactions
references/rust.md	Rust — Cargo setup, connection, query, Value types
references/go.md	Go — module setup, connection, query, transactions
references/swift.md	Swift — SPM setup, connection, query, async/await
references/import.md	COPY FROM, LOAD FROM, DataFrame import, cloud storage, performance tips
references/export.md	COPY TO, DataFrame export (pandas/polars/arrow), DuckDB export
references/graph-algorithms.md	PageRank, Louvain, WCC, SCC, K-Core, shortest paths — PROJECT_GRAPH
references/vector-search.md	HNSW index, CREATE/QUERY/DROP_VECTOR_INDEX, RAG pattern
references/full-text-search.md	BM25, CREATE/QUERY/DROP_FTS_INDEX, stemmers
references/llm-embeddings.md	CREATE_EMBEDDING — OpenAI, Ollama, Google, Bedrock, Voyage AI
references/attach.md	ATTACH/DETACH — PostgreSQL, DuckDB, SQLite, Delta Lake, Iceberg, Neo4j
references/cli.md	lbug shell flags, commands, output modes, batch/scripting mode
references/explorer.md	Ladybug Explorer Docker GUI — launch, env vars, volume mount

Common task routing

Task	Read
Schema design, Cypher queries, differences from Neo4j	cypher-reference.md
Python integration	python.md
Node.js / TypeScript integration	nodejs.md
Java integration	java.md
Rust integration	rust.md
Go integration	go.md
Swift / iOS / macOS integration	swift.md
Bulk import — COPY FROM, LOAD FROM, DataFrames, cloud storage	import.md
Bulk export — COPY TO, DataFrame export, DuckDB	export.md
PageRank, Louvain, WCC, SCC, K-Core, shortest paths	graph-algorithms.md
HNSW vector similarity search, RAG	vector-search.md
Full-text search (BM25)	full-text-search.md
LLM embeddings (OpenAI, Ollama, Bedrock…)	llm-embeddings.md
ATTACH to PostgreSQL, DuckDB, Delta Lake, Neo4j	attach.md
CLI shell, batch scripts	cli.md
Ladybug Explorer browser GUI (Docker)	explorer.md

Key gotchas

1. SET n.prop = NULL to remove a property (not REMOVE)
2. label(n) not labels(n); id(n) not elementId(n)
3. UNWIND instead of FOREACH
4. List functions use list_ prefix: list_concat, list_sort, etc.
5. Variable-length paths must have an upper bound (default 30): [:Follows*1..5]
6. LOAD FROM (not LOAD CSV FROM) — supports CSV, Parquet, JSON, DataFrames
7. No manual index creation — primary key index is automatic; use FTS/vector extensions for search