langchain-age 0.1.2

LangChain integration for Apache AGE (graph) + pgvector (vector) on PostgreSQL

langchain-age

LangChain integration for Apache AGE (graph) + pgvector (vector) on PostgreSQL.

Drop-in replacement for langchain-neo4j — same API, runs on PostgreSQL instead of Neo4j.

from langchain_age import AGEGraph, AGEVector, AGEGraphCypherQAChain

v0.1.1 — Initial stable release. The public API (AGEGraph, AGEVector, AGEGraphCypherQAChain) is considered stable. Breaking changes in 0.1.x will be avoided where possible.

Tested with

• Python 3.10–3.14
• PostgreSQL 18 + Apache AGE 1.7.0 + pgvector 0.8.2

Known limitations

• AGE does not support parameterised Cypher ($param) — langchain-age provides mogrify-based safe value escaping as a workaround
• Async methods use run_in_executor wrapping (not native psycopg.AsyncConnection yet)

Installation

Three install modes depending on what you need:

# Graph only
pip install langchain-age

# Graph + Vector (everything)
pip install "langchain-age[all]"

The Apache AGE Python driver is vendored — no separate installation required. Graph features work out of the box with pip install langchain-age.

Quick Start

1. Start the database

cd docker
docker compose up -d

Single container: PostgreSQL 18 + Apache AGE 1.7.0 + pgvector.

2. Graph mode

from langchain_age import AGEGraph

graph = AGEGraph(
    "host=localhost port=5433 dbname=langchain_age user=langchain password=langchain",
    graph_name="my_graph",
)

graph.query("CREATE (:Person {name: 'Alice', age: 30})")
results = graph.query("MATCH (n:Person) RETURN n.name AS name")
# [{'name': 'Alice'}]

3. Vector mode

from langchain_age import AGEVector, DistanceStrategy

store = AGEVector(
    connection_string="host=localhost port=5433 dbname=langchain_age user=langchain password=langchain",
    embedding_function=my_embeddings,
    collection_name="docs",
    distance_strategy=DistanceStrategy.COSINE,
)

store.add_texts(["Apache AGE adds Cypher to PostgreSQL."])
results = store.similarity_search("graph database", k=5)

4. Graph + Vector (GraphRAG)

# Vectorise existing graph nodes
store = AGEVector.from_existing_graph(
    embedding=my_embeddings,
    connection_string="...",
    graph_name="my_graph",
    node_label="Document",
    text_node_properties=["title", "content"],
)

5. Cypher QA Chain

from langchain_age import AGEGraph, AGEGraphCypherQAChain
from langchain_openai import ChatOpenAI

chain = AGEGraphCypherQAChain.from_llm(
    ChatOpenAI(model="gpt-4o-mini"),
    graph=AGEGraph("...", "movies"),
    allow_dangerous_requests=True,
)
answer = chain.run("Which movies did Tom Hanks act in?")

6. Long-term Memory & Checkpointing

Uses the same PostgreSQL instance via langgraph-checkpoint-postgres:

from langgraph.store.postgres import PostgresStore

with PostgresStore.from_conn_string("host=localhost port=5433 ...") as store:
    store.setup()
    store.put(("users", "123"), "prefs", {"theme": "dark"})

AGE graph tables, pgvector tables, and LangGraph store tables coexist in the same database.

Features

Component	Class	Description
Graph	AGEGraph	GraphStore backed by Apache AGE. Cypher execution, schema introspection, GraphDocument upserts.
Vector	AGEVector	VectorStore backed by pgvector. Cosine/L2/IP distance, HNSW & IVFFlat indexes, hybrid search (vector + full-text via RRF), MMR.
Chain	AGEGraphCypherQAChain	LLM generates Cypher, executes against AGE, returns natural-language answer.

Security

• SQL identifier validation at construction (validate_sql_identifier)
• Cypher identifier backtick-quoting for all 25 AGE reserved words (escape_cypher_identifier)
• OpenCypher-standard string escaping with '' doubling (escape_cypher_string)
• allow_dangerous_requests gate on the QA chain
• Double-quoted SQL table/index names throughout

langchain-neo4j API Compatibility

Feature	langchain-neo4j	langchain-age
from_existing_graph()	Neo4jVector.from_existing_graph()	AGEVector.from_existing_graph()
from_existing_index()	Neo4jVector.from_existing_index()	AGEVector.from_existing_index()
similarity_search_with_relevance_scores()	Yes	Yes
as_retriever()	Yes	Yes
Hybrid search	Lucene full-text	PostgreSQL tsvector + RRF
include_types / exclude_types	Yes	Yes
add_graph_documents()	Yes	Yes
Context manager	Yes	Yes
Batch insert	UNWIND ... IN TRANSACTIONS OF 1000 ROWS	executemany with batch_size=1000

AGE vs Neo4j

	Neo4j	Apache AGE
Cypher execution	Bolt protocol	SQL-wrapped: SELECT * FROM cypher(...)
Connection	neo4j://	PostgreSQL DSN
Vector search	Native index	pgvector extension
APOC	Available	Not available
Data types	Native graph	agtype (JSON superset)
Parameterised Cypher	Yes ($param)	mogrify-based (%s placeholders)

langchain-age handles SQL wrapping automatically — you write plain Cypher.

Documentation

Language	Getting Started	Tutorial	API Reference
English	getting-started.md	tutorial.md	api-reference.md
Korean	getting-started.md	tutorial.md	api-reference.md

Notebooks

Notebook	Description
01_graph.ipynb	AGEGraph: Cypher CRUD, schema, GraphDocument
02_vector.ipynb	AGEVector: similarity, hybrid, MMR, filters, HNSW
03_graph_vector.ipynb	GraphRAG: from_existing_graph, QA chain

01 requires no API key. 02–03 use OpenAI via getpass.

Running Tests

# Unit tests (no DB required) — 65 tests
pytest tests/unit/

# Integration tests (requires Docker container) — 53 tests
export LANGCHAIN_AGE_TEST_DSN="host=localhost port=5433 dbname=langchain_age user=langchain password=langchain"
pytest tests/integration/

Project Structure

langchain-age/
├── docker/                          # PG18 + AGE + pgvector container
│   ├── Dockerfile
│   ├── docker-compose.yml
│   └── init/01_init_extensions.sql
├── langchain_age/
│   ├── __init__.py                  # Lazy imports (3-mode support)
│   ├── graphs/
│   │   └── age_graph.py             # AGEGraph (GraphStore)
│   ├── vectorstores/
│   │   └── age_vector.py            # AGEVector (VectorStore)
│   ├── chains/
│   │   └── graph_cypher_qa_chain.py # AGEGraphCypherQAChain
│   └── utils/
│       ├── agtype.py                # Vertex/Edge/Path → dict conversion
│       └── cypher.py                # SQL wrapping, escaping, validation
├── tests/
│   ├── conftest.py                  # Auto-skip integration when DSN unset
│   ├── unit/                        # 65 tests, no DB
│   └── integration/                 # 53 tests, live DB
├── pyproject.toml
├── LICENSE                          # MIT
├── CHANGELOG.md                     # Version history
└── .github/workflows/ci.yml        # Lint + unit (3.10–3.13) + integration

Python Support

Tested on Python 3.10, 3.11, 3.12, 3.13, 3.14.

License

MIT