PGVector AssertionError('_async_engine not found') with a RAG tool

[cj2001]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

Creation of vectors

import requests
import os

from bs4 import BeautifulSoup
import html2text
from dotenv import load_dotenv

from langchain_community.embeddings import OpenAIEmbeddings
from langchain.text_splitter import RecursiveCharacterTextSplitter, MarkdownHeaderTextSplitter
from langchain_postgres.vectorstores import PGVector

from settings import Settings


PGDATABASE = '...'
CONNECTION_STRING = f"postgresql+psycopg://{Settings.PGUSER}:{Settings.PGPASSWORD}@{Settings.PGHOST}:{Settings.PGPORT}/{PGDATABASE}"
COLLECTION_NAME = "..."

...a bunch of markdown webpages read...

text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=3000,
    chunk_overlap=150,
    length_function=len,
    is_separator_regex=False,
)

text_splits = text_splitter.create_documents([text_content])

docs = md_header_splits + text_splits

embedding = OpenAIEmbeddings(openai_api_key=Settings.OPENAI_API_KEY)

vector_store = PGVector(
    embeddings=embedding,
    collection_name=COLLECTION_NAME,
    connection=CONNECTION_STRING,
    pre_delete_collection=True,
    use_jsonb=True,
)

document_ids = vector_store.add_documents(documents=docs)

print(document_ids[0:5])

Retriever code for these vectors:

from langchain.tools.retriever import create_retriever_tool
from langchain_postgres.vectorstores import PGVector
from langchain_openai import OpenAIEmbeddings

embeddings = OpenAIEmbeddings()

COLLECTION_NAME = "..."
PGDATABASE = '...'
CONNECTION_STRING = f"postgresql+psycopg://{settings.PGUSER}:{settings.PGPASSWORD}@{settings.PGHOST}:{settings.PGPORT}/{PGDATABASE}"

store = PGVector(
    collection_name=COLLECTION_NAME,
    connection=CONNECTION_STRING,
    embeddings=embeddings,
)

retriever = store.as_retriever(search_kwargs={"k": 5})

MyDocs = create_retriever_tool(
    retriever,
    "...",
    """
    Description of RAG.
    """,
)

Description

Hi, all.

I have created vector embeddings using PGVector following the docs (see the first code snippet). I can see that this creates tables in postgres with embeddings. I then created a retriever as a tool in LangGraph with code that is the second snippet. However, when I ask a question in LangGraph that would use this RAG, I get the following error:

Error: AssertionError('_async_engine not found') Please fix your mistakes.

It seems this is related to PGVector and not LangGraph. Any tips?

System Info

langchain 0.3.14
langchain-community 0.3.14
langchain-core 0.3.29
langchain-openai 0.2.14
langchain-postgres 0.0.13
langchain-text-splitters 0.3.5
langgraph 0.2.61
langgraph-api 0.0.15
langgraph-checkpoint 2.0.9
langgraph-checkpoint-postgres 2.0.11
langgraph-cli 0.1.65
langgraph-sdk 0.1.48
langsmith 0.2.10

[cj2001]

I am going to close this discussion because I found the answer elsewhere. However, for the sake of documenting the solution, the answer came from this issue.

The bottom line is that the retriever needs an async engine. Just passing the connection string is not sufficient. So the change in the retriever tool looks like this:

from langchain.tools.retriever import create_retriever_tool
from langchain_postgres.vectorstores import PGVector
from langchain_openai import OpenAIEmbeddings

from sqlalchemy.ext.asyncio import create_async_engine

embeddings = OpenAIEmbeddings()

COLLECTION_NAME = "..."
PGDATABASE = '...'
CONNECTION_STRING = f"postgresql+psycopg://{settings.PGUSER}:{settings.PGPASSWORD}@{settings.PGHOST}:{settings.PGPORT}/{PGDATABASE}"

engine = create_async_engine(CONNECTION_STRING)

store = PGVector(
    collection_name=COLLECTION_NAME,
    connection=engine,
    embeddings=embeddings,
)

retriever = store.as_retriever(search_kwargs={"k": 5})

MyDocs = create_retriever_tool(
    retriever,
    "...",
    """
    Description of RAG.
    """,
)

(Note the use of the sqlalchemy package, creation of engine, and then using engine in PGVector.)