Apache AGE

Quick Start

CREATE EXTENSION IF NOT EXISTS age;
LOAD 'age';
SET search_path = ag_catalog, "$user", public;
SELECT create_graph('my_graph');

SELECT * FROM cypher('my_graph', $$
  CREATE (a:Person {name: 'Alice'})-[:KNOWS]->(b:Person {name: 'Bob'})
  RETURN a, b
$$) AS (a agtype, b agtype);

Reference Loading

Principle	Reference	When to Load
The Wrapper Contract	the-wrapper-contract.md	MANDATORY for first-time AGE users. Load for cypher() mechanics, search_path, parameters, unsupported features
agtype Is Everything	agtype-is-everything.md	Load when casting agtype to SQL types, debugging type errors, or handling nulls
Hybrid Is the Point	hybrid-is-the-point.md	MANDATORY before writing any SQL+Cypher query. CTEs, JOINs, mutation guards, multi-graph
Schema Is Storage	schema-is-storage.md	Load when designing graph schema, choosing labels/properties, or modeling relationships
Nothing Is Automatic	nothing-is-automatic.md	MANDATORY before performance work. Indexes, EXPLAIN, batch loading, when to use SQL instead

Do NOT load:

• schema-is-storage.md for pure query questions (no schema design involved)
• nothing-is-automatic.md for schema design unless indexing is the concern
• the-wrapper-contract.md for users already familiar with cypher() mechanics

Freedom calibration:

• Low freedom (follow exactly): cypher() wrapper syntax, AS column lists, search_path, index DDL, mutation CTE guards
• Medium freedom (adapt to context): graph modeling, hybrid query composition, performance tuning
• High freedom (choose freely): property naming, schema partitioning, application-layer validation

Think in AGE

• Every Cypher query is a SQL function call — plan the SQL wrapper first, then the Cypher inside it
• agtype is the only type that crosses the boundary — cast explicitly at every SQL touchpoint
• Graphs are PostgreSQL schemas — labels are tables, vertices and edges are rows with agtype property blobs
• Nothing exists until you create it — no default indexes, no property constraints, no type enforcement
• Hybrid is the value proposition — when a query is pure aggregation or pure OLTP, drop to SQL (it's ~15x faster for aggregation per Issue #2194)

Critical Gotchas

1. Always define a column list

cypher() returns SETOF record — PostgreSQL requires an explicit column list:

SELECT * FROM cypher('g', $$ MATCH (n) RETURN n $$) AS (n agtype);

2. Terminal clauses need a dummy column

CREATE, SET, DELETE at the end of a query return no rows, but the column list is still mandatory:

SELECT * FROM cypher('g', $$
  CREATE (:Person {name: 'Alice'})
$$) AS (v agtype);
-- Returns 0 rows

3. SET = replaces all properties; use += to merge

-- Replaces ALL properties with just {age: 30}
SET n = {age: 30}

-- Merges {age: 30} into existing properties
SET n += {age: 30}

4. CREATE/SET/REMOVE not allowed in JOINs

Mutations inside a JOIN cause errors. Wrap in a CTE instead:

WITH new_vertex AS (
  SELECT * FROM cypher('g', $$ CREATE (p:Person {name: 'Alice'}) RETURN p $$) AS (v agtype)
)
SELECT * FROM new_vertex;

5. search_path must include ag_catalog

SET search_path = ag_catalog, "$user", public;
-- Or fully qualify: SELECT * FROM ag_catalog.cypher(...)

NEVER

• NEVER omit the AS column list — cypher() returns SETOF record, PostgreSQL demands it
• NEVER use SET n = {...} when you mean SET n += {...} — = replaces ALL properties silently
• NEVER put CREATE/SET/REMOVE inside a JOIN — wrap in CTE; mutations break PostgreSQL transaction handling
• NEVER use unbounded [:REL*] — paths explode at scale (7s → 7min on 1.5M vertices, Issue #195)
• NEVER assume indexes exist — AGE creates none by default; use EXPLAIN inside Cypher to verify
• NEVER compare agtype to SQL types without casting — agtype can only cast to varchar among string types (not json/text, Issues #1225, #1996)
• NEVER query without a label — scans entire _ag_label_vertex parent table
• NEVER use multiple labels on one vertex — AGE supports exactly zero or one (by design, Discussion #109)
• NEVER build maps with 50+ fields — agtype_build_map() hits the 100-argument PostgreSQL function limit (Issue #1840)
• NEVER pass null to agtype functions without checking — some null inputs crash the server (Issues #1956, #1303)

Common Recipes

1-hop neighbors with SQL JOIN

SELECT u.email, friend.name::varchar AS friend_name
FROM users u
JOIN (
  SELECT * FROM cypher('social', $$
    MATCH (p:Person)-[:KNOWS]->(f:Person)
    WHERE p.userId = 1042
    RETURN p.userId, f.name
  $$) AS (user_id agtype, name agtype)
) g ON u.id = g.user_id::int;

MERGE upsert

SELECT * FROM cypher('g', $$
  MERGE (p:Person {email: 'alice@example.com'})
  ON CREATE SET p.created_at = timestamp(), p.login_count = 1
  ON MATCH SET p.login_count = p.login_count + 1
  RETURN p
$$) AS (p agtype);

UNWIND bulk create

SELECT * FROM cypher('g', $$
  UNWIND [{name: 'Alice', age: 30}, {name: 'Bob', age: 25}] AS attrs
  CREATE (p:Person {name: attrs.name, age: attrs.age})
  RETURN p.name
$$) AS (name agtype);

Bounded path search

SELECT * FROM cypher('g', $$
  MATCH (a:Person {name: 'Alice'})-[:KNOWS*1..3]->(b:Person)
  RETURN DISTINCT b.name
  LIMIT 100
$$) AS (name agtype);