Skip to main content


This template performs RAG using the self-query retrieval technique. The main idea is to let an LLM convert unstructured queries into structured queries. See the docs for more on how this works.

Environment Setup​

In this template we'll use OpenAI models and an Elasticsearch vector store, but the approach generalizes to all LLMs/ChatModels and a number of vector stores.

Set the OPENAI_API_KEY environment variable to access the OpenAI models.

To connect to your Elasticsearch instance, use the following environment variables:


For local development with Docker, use:

export ES_URL = "http://localhost:9200"
docker run -p 9200:9200 -e "discovery.type=single-node" -e "" -e ""


To use this package, you should first have the LangChain CLI installed:

pip install -U "langchain-cli[serve]"

To create a new LangChain project and install this as the only package, you can do:

langchain app new my-app --package rag-self-query

If you want to add this to an existing project, you can just run:

langchain app add rag-self-query

And add the following code to your file:

from rag_self_query import chain

add_routes(app, chain, path="/rag-elasticsearch")

To populate the vector store with the sample data, from the root of the directory run:


(Optional) Let's now configure LangSmith. LangSmith will help us trace, monitor and debug LangChain applications. You can sign up for LangSmith here. If you don't have access, you can skip this section

export LANGCHAIN_API_KEY=<your-api-key>
export LANGCHAIN_PROJECT=<your-project> # if not specified, defaults to "default"

If you are inside this directory, then you can spin up a LangServe instance directly by:

langchain serve

This will start the FastAPI app with a server is running locally at http://localhost:8000

We can see all templates at We can access the playground at

We can access the template from code with:

from langserve.client import RemoteRunnable

runnable = RemoteRunnable("http://localhost:8000/rag-self-query")

Help us out by providing feedback on this documentation page: