# Installation

## Requirements

- **Python** 3.11–3.14. On Python 3.10, pip installs 0.2.0 instead, without an error, and
  these pages do not describe it.
- A running **AgensGraph** 2.17 or later. An older server is refused at connect.
- The **pgvector** extension, for the vector store and `AgensStore` semantic search

Every database call goes through the
[`agensgraph-python`](https://github.com/skaiworldwide-oss/agensgraph-python) 2.0 driver
(psycopg 3), which is installed with the package. The integration targets `langchain-core` 1.x
and `langgraph` 1.x, and has **no dependency on the archived `langchain-community`**.

## Install

```bash
pip install -U langchain-agensgraph
```

or through the `agensgraph-ai` umbrella package, which installs the same thing:

```bash
pip install "agensgraph-ai[langchain]"
```

There is no `agensgraph_ai` module; either way you import `langchain_agensgraph`. This
documentation covers **0.3.1**. 0.3.1 is a documentation release; its code is 0.3.0's.

The examples that embed text or call an LLM use `langchain-openai`, which is not a
dependency. Install it separately if you want to run them as written:

```bash
pip install langchain-openai
```

## Set up pgvector

AgensGraph does not bundle the `pgvector` extension. Build and install it against your
AgensGraph install's `pg_config`:

```bash
git clone https://github.com/pgvector/pgvector.git
export PG_CONFIG=/path/to/agens/bin/pg_config
cd pgvector && make && make install
```

```sql
-- in your AgensGraph database, as a superuser:
CREATE EXTENSION IF NOT EXISTS vector;
```

On a large graph, enabling `auto_gather_graphmeta` on the database
(`ALTER DATABASE mydb SET auto_gather_graphmeta = on`) makes `DETACH DELETE` much faster.

## Connecting

`AgensGraph`, `AgensChatMessageHistory`, `AgensSaver`, and `AgensStore` accept a `conf`
dictionary of `psycopg.connect` keyword arguments. The last three can instead take an existing
`AgensGraph` as `graph=`:

```python
conf = {
    "dbname": "mydb",
    "user": "postgres",
    "password": "postgres",
    "host": "localhost",
    "port": 55432,
}
```

The vector store takes a libpq `url` instead, such as
`postgresql://user:password@localhost:55432/mydb`, or reads it from the `AGENSGRAPH_URL`
environment variable. It can also be built on an existing `AgensGraph` (`graph=`). To share
one connection pool between the graph store and several vector stores, create an
`AgensEngine` and pass it as `engine=`. Pass `create=True` to have the graph store create the
graph if it does not exist. See [Usage](usage.md) for full examples.