agensgraph

A high-performance graph database built on PostgreSQL.
Graphs and tables in one database. SQL, openCypher and ISO GQL in one query.
AgensGraph is developed by SKAI Worldwide, formerly Bitnine.

Documentation · Quick start · AI and agents · Discussions · Releases

What AgensGraph is

• PostgreSQL with the property graph model built in. A fork of PostgreSQL, not an extension.
• A graph is a schema, a label is a table, a vertex or an edge is a row.
• Graph data and relational tables live in one database and commit in one transaction.
• Query with SQL, openCypher and ISO GQL (ISO/IEC 39075). Mix them in one statement.
• Everything PostgreSQL has works on graph data: transactions, roles, row-level security, replication, pg_dump, EXPLAIN, extensions such as pgvector.
• Apache License 2.0.

Quick start

1. Start a server

docker run --name agensgraph -e POSTGRES_PASSWORD=agensgraph -d skaiworldwide/agensgraph:latest
docker exec -it agensgraph agens -U agens -d postgres

2. Create a graph

CREATE GRAPH movies;
SET graph_path = movies;

CREATE (keanu:person {name: 'Keanu Reeves'}),
       (carrie:person {name: 'Carrie-Anne Moss'}),
       (matrix:movie {title: 'The Matrix', year: 1999}),
       (wick:movie {title: 'John Wick', year: 2014}),
       (keanu)-[:acted_in {role: 'Neo'}]->(matrix),
       (carrie)-[:acted_in {role: 'Trinity'}]->(matrix),
       (keanu)-[:acted_in {role: 'John Wick'}]->(wick);

3. Match a pattern

MATCH (:person {name: 'Carrie-Anne Moss'})-[:acted_in]->(m:movie)<-[:acted_in]-(co:person)
RETURN co.name AS co_star, m.title AS movie;

    co_star     |    movie
----------------+--------------
 "Keanu Reeves" | "The Matrix"
(1 row)

4. Follow a path of any length

MATCH (:person {name: 'Carrie-Anne Moss'})-[:acted_in*1..3]-(m:movie)
RETURN DISTINCT m.title AS reachable ORDER BY reachable;

  reachable
--------------
 "John Wick"
 "The Matrix"
(2 rows)

5. Join the graph to a table

CREATE TABLE my_ratings (title text, stars int);
INSERT INTO my_ratings VALUES ('The Matrix', 5), ('John Wick', 4);

SELECT f.title, r.stars
FROM (MATCH (:person {name: 'Keanu Reeves'})-[:acted_in]->(m:movie)
      RETURN m.title AS title) AS f
JOIN my_ratings r ON r.title = f.title #>> '{}'
ORDER BY r.stars DESC;

    title     | stars
--------------+-------
 "The Matrix" |     5
 "John Wick"  |     4
(2 rows)

Next: the example dataset · combining SQL and Cypher · paths

Query languages

Language	Use it for	Docs
SQL	Tables, DDL, administration, and queries over a Cypher subquery	Combining SQL and Cypher
openCypher	Patterns, paths, graph writes	Reading, Writing
ISO GQL	LET, FILTER, FOR, NEXT, FINISH, INSERT, CALL { }, subquery expressions	GQL conformance

GQL and Cypher are one language. No mode switch:

MATCH (p:person)-[:acted_in]->(m:movie)
RETURN p.name AS actor, count(m) AS movies
NEXT FILTER movies > 1
RETURN actor;

     actor
----------------
 "Keanu Reeves"
(1 row)

Features

Area	What you get
Graph model	Vertex and edge labels, label inheritance, property indexes, constraints
Paths	Variable-length patterns, shortestpath(), allshortestpaths(), dijkstra()
Subqueries	CALL { }, OPTIONAL CALL, EXISTS { }, COUNT { }, COLLECT { }, VALUE { }
Typed columns	Store chosen properties in typed, indexable columns (typed columns)
Search	Vector and full-text search on properties through PostgreSQL extensions (vector, full text)
Planner	Row estimates for paths, label pruning, skipped scans for unused endpoints
Security	Grants and row-level security on labels, enforced for Cypher reads and writes
Operations	pg_dump / pg_restore, pg_upgrade, streaming and logical replication

AI and agents

• Knowledge graph, embeddings (with pgvector) and agent memory in one PostgreSQL database.
• GraphRAG in one statement: vector, full-text or hybrid search, then a graph traversal.
• MCP servers give AI assistants and agents tool access to the graph.

Integration	What it provides	Package
MCP: Cypher	Schema introspection and read/write Cypher tools for MCP clients such as Claude Desktop	mcp-agensgraph-cypher
MCP: Memory	Knowledge-graph memory for agents: entities, relations, observations	mcp-agensgraph-memory
MCP: Data modeling	Graph model validation; generated ingest and constraint Cypher	mcp-agensgraph-data-modeling
LangChain	Graph store, pgvector vector store, chat history, LangGraph checkpointer, LLM graph transformer	langchain-agensgraph
LlamaIndex	PropertyGraphIndex and VectorStoreIndex stores, hybrid search, Text2Cypher	llama-index-agensgraph
LightRAG	All four LightRAG storage roles in one database, with workspace multi-tenancy	lightrag-agensgraph
cognee	Graph and vector adapters for the cognee AI-memory framework	cognee-agensgraph

pip install "agensgraph-ai[all]"     # or one of: [mcp] [langchain] [llama-index] [lightrag] [cognee]

AgensGraph AI documentation · source and demos · GraphRAG without a framework

Install

Route	How
Docker	skaiworldwide/agensgraph:v<release>-<base>. Bases: bookworm, trixie, alpine3.23, alpine3.24, rocky9, rocky10 (Docker Hub)
Source	Build from source
Upgrade from 2.17	Upgrading to 2.18

Drivers

Any PostgreSQL driver connects (drivers and client libraries). The Python driver adds graph types:

Language	Package	Docs	Repository
Python	agensgraph-python	AgensGraph for Python	agensgraph-python

Documentation

• Manual
• Tuning queries
• Configuration
• Release notes and downloads
• In-tree reference: doc/src/sgml/agensgraph.sgml and doc/src/sgml/ref/cypher_*.sgml

Community

• Questions: Discussions
• Bugs and feature requests: Issues
• Code: pull requests against main

License

• AgensGraph: Apache License 2.0. See NOTICE.
• Portions from PostgreSQL: PostgreSQL License. See COPYRIGHT.