AgensGraph AI Tools and Libraries#

Welcome to AgensGraph AI — a collection of integrations that connect popular AI and LLM frameworks to AgensGraph, Skai’s PostgreSQL-based multi-model graph database. Each integration lets you keep your knowledge graphs and your vector embeddings in a single AgensGraph instance, queried with Cypher over the standard PostgreSQL wire protocol.

Libraries at a glance#

Library

What it gives you

LangChain

Graph store, vector store, chat-message history, a LangGraph checkpointer, and an LLM graph transformer for LangChain 1.x.

LlamaIndex

PropertyGraphIndex and VectorStoreIndex stores, with hybrid search, metadata filtering, and AgensGraph-tuned Text2Cypher.

LightRAG

All four LightRAG storage roles on one database, with workspace multi-tenancy.

cognee

Graph and vector adapters for the cognee AI-memory framework.

MCP servers

Three MCP servers — Cypher, memory, and data modeling — that expose AgensGraph to clients like Claude Desktop.

Installing#

Each integration is released on PyPI on its own, and the agensgraph-ai package installs any combination of them under one name. Name the ones you need as extras:

pip install "agensgraph-ai[langchain]"
pip install "agensgraph-ai[langchain,lightrag]"
pip install "agensgraph-ai[all]"

Extra

Installs

Then import

langchain

langchain-agensgraph

langchain_agensgraph

llama-index

llama-index-agensgraph

llama_index_agensgraph

lightrag

lightrag-agensgraph

lightrag_agensgraph

cognee

cognee-agensgraph

cognee_agensgraph

mcp

mcp-agensgraph-cypher, mcp-agensgraph-memory, mcp-agensgraph-data-modeling

run as commands

all

all of the above

  • Name at least one extra. pip install agensgraph-ai on its own installs nothing, and a misspelled extra installs nothing either: pip and uv only print a warning and still succeed.

  • There is no agensgraph_ai module. Each extra is a shortcut to the real package, which you import under its own name, and pip install langchain-agensgraph produces the same environment as pip install "agensgraph-ai[langchain]".

  • The mcp extra is for hosting a server yourself over HTTP or SSE. An MCP client normally launches the servers with uvx, which needs no install (see Installation).

  • Python 3.11 or later; the LightRAG and cognee integrations (and so [all]) need 3.11 to 3.13.

A common foundation#

All of these integrations share the same underlying model:

  • One database for graph and vectors. They talk to AgensGraph through the agensgraph-python 2.0 driver, which runs on psycopg 3, and run Cypher inside it, so a single instance can hold both your property graph and your embeddings.

  • pgvector for embeddings. Vector search uses the pgvector extension. AgensGraph does not bundle it, so you install it once into your database (see each library’s Installation page).

  • AgensGraph 2.17 or later is required. The driver refuses an older server at connect. SHOW agversion tells you which you have; from 2.18 a release reports four numbers, such as 2.18.6.0, and the first two are the line.

Choosing an integration#

  • Already building in LangChain or LlamaIndex? Use the matching integration — it plugs into that framework’s standard graph-store / vector-store interfaces.

  • Want a batteries-included graph-RAG or memory pipeline? LightRAG and cognee handle ingestion, extraction, and retrieval for you.

  • Want an LLM agent (e.g. Claude Desktop) to use AgensGraph directly through tools, with no Python glue? Use the MCP servers.

Note

The source for every integration lives in the agensgraph-ai repository, and each library ships a runnable demo suite with pre-executed notebooks. Links to those demos appear on each library’s Usage page.