LangChain Templates: A powerful collection of customizable chains and agents

Editing and customizing the internals of LangChain agents and chains is not a pleasant experience. LangChain Templates solves this issue by exposing the inner workings as downloadable templates.

LangChain Templates: A powerful collection of customizable chains and agents
LangChain Templates: An Easy Way to Customize Chains and Agents
New to LangChain? Start with this introductory post first. It'll give you a great overview of everything you need to know before diving in. Come back when you're done!

Introduction to LangChain

Since I've been covering LangChain recently, I am not going to write yet another introduction about the framework.

If you're not familiar with LangChain, fear not! This blog is just for you. Go ahead and start here and come back when you're done.

I'll wait for you to finish. โณ ๐Ÿ‘€

Prompt Templates... NOT

Phew! Thanks to the posts here, you're now a LangChain Master, and we can move on. LangChain provides developers with Prompt Templates that act as placeholders allowing us to define LLM prompts quickly and easily.

Here's how a PromptTemplate works:

from langchain.prompts import PromptTemplate

prompt_template = PromptTemplate.from_template(
    "Tell me a {adjective} joke about {content}."
prompt_template.format(adjective="funny", content="chickens")


But LangChain already has Prompt Templates you say! So how are LangChain Templates different? The similar names may confuse you but as we'll see in this post, they're not the same thing.

Here's why ๐Ÿ‘‡

Exploring LangChain Templates

From the official docs:

LangChain Templates are the easiest and fastest way to build a production-ready LLM application. These templates serve as a set of reference architectures for a wide variety of popular LLM use cases. They are all in a standard format which make it easy to deploy them with LangServe.

As Harrison Chase, the co-founder of LangChain wrote in this ProductHunt comment, the need developers have to edit and customize the internals of the chains/agents prompted the creation of LangChain Templates.

Instead of having all the chains/agents as part of the Python library's source code, LangChain Templates now exposes all the inner workings of the relevant chains and agents as downloadable templates easily accessible directly within the application code.

Let's see how this works with the rag-conversation template one of the most popular LLM use-cases.

RAG Conversation Template

We're going to write just a few necessary lines of code for the example below, then, we'll take a look at how we can access the templates from within your project.

Let's get started! We'll need to do the following:

  • Install the LangChain CLI
  • Add a rag-conversation Template to a new (or existing) Project
  • Review Directory Structure
  • Deploy and Test (Optional)
    • OpenAI and Pinecone API Keys required

Install the LangChain CLI

In your terminal window, type the following and hit return to install the LangChain CLI:

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

Create a New LangChain Project

In your work directory, and using the newly installed LangChain CLI, let's use it to create our first app and include the rag-conversation package by specifying it using the --package flag, as shown below:

langchain app new testApp --package rag-conversation
You may need to confirm twice to proceed with the installation. The first time the CLI will ask you to proceed with installing the Template, and the second time it will ask you if you'd like to generate route code for these packages. Choose Yes for both.

After the new project is created, we can navigate to the directory testApp. On your terminal type cd testApp and hit return.

You'll find the following directories in your project root:

  • /app: This directory contains the base code that defines a FastAPI that will serve your application code.
  • /packages: This directory will contain all LangChain Templates that you opt to install via the langchain app new testApp --package <template-name> command.

Since we've installed the rag-conversation template we find it in the directory as expected:

LangChain Template: rag-conversation

That's all we need to do, we can just navigate to the nested rag-conversation directory where we'll find the file (as shown above).

This file includes all of the required code to run this chain. That's where we can easily modify and customize the behaviour based on our application's specific needs.

Add to Existing LangChain Project

Alternatively, if you have an existing project and you'd like to install one or more LangChain Templates, you could use the following CLI command:

langchain app add rag-conversation

Running the LangChain Template (Optional)

โš ๏ธ
This template uses OpenAI and Pinecone Vector DBMS by default, so you'll need the appropriate keys to run and test it. You can alternatively modify the code to run based on your setup.

To run the template, we'll just need to import our chain and register the FastAPI route so we can test it (below). Here's how we can do this:

Go to and import the chain by adding the following:

from rag_conversation import chain as rag_conversation_chain

Then, simply register the FastAPI route by adding the following:

add_routes(app, rag_conversation_chain, path="/rag_conversation")

Setting OpenAI Key

Let's go ahead and add our OpenAI API Key by typing the following into our terminal: export OPENAI_API_KEY="YOUR_API_KEY_HERE"

If you don't have an OpenAI Key, you'll need to get your credit card out and sign into your OpenAI account to generate a key.

Pinecone Set Up

If you don't have a Pinecone account, you can create a free one here. Let's go ahead and create our index.

Creating Pinecone Index

From your Pinecone account, go to the Indexes tab and click on 'Create Index'. You can use the following settings:

Pinecone LangChain Template Demo

Make sure the index name here langchain-test matches the one specified on line 26 in the file. It is langchain-test by default so you can skip this step if you haven't customized the name.

Pinecone API Keys

From your Pinecone account, go to the Indexes tab and click on 'Create Index'. You can use the following settings:

To add your Pinecone API Keys we'll need to type the following into our terminal:

# Grab your Pinecone API Key

# Modify to match your Pinecone environment name

Pinecone Project Settings

Note: The environment name is visible in the Indexes tab for each index from the list.

Running the Project

This step is important, the first time you run your application, we'll need to add the data from the demo file to the Pinecone index. To do so, uncomment lines 29 to 43 in your file. It's safe to comment them back after the initial run.

LangChain CLI

Finally, to test the template, we can use LangServe and launch the project by typing the following in the terminal: langchain serve and hit return:

Bam! ๐ŸŽ‰๐ŸŽ‰๐ŸŽ‰

LangServe Startup Success

Swagger Docs

Now if you head over to on your machine, you should be able to see the FastAPI generated Swagger docs for the LangChain Template we just set up!

โš ๏ธ
Make sure the address and port match the FastAPI application running on your local machine.

Testing the Template via Jupyter Notebook

Super, to quickly test our LangChain Template via the RESTful API we just launched we'll need Jupyter Notebook.

Go ahead and launch Jupyter Notebook by typing the following into your terminal window and then hitting return:

jupyter notebook

Once the notebook loads up and the kernel is ready, copy and paste the below code into the first cell and hit run (or CMD+Return):

from langserve.client import RemoteRunnable

rag_app = RemoteRunnable("")

question = "How does agent memory work?"
answer = rag_app.invoke(
        "question": question,
        "chat_history": [],


The application should respond with something similar to the below:

The agent memory consists of two components: the memory stream and the retrieval model. (Etc...)

Of course, this is based on the vector embeddings from the ingested demo URL provided in the LangChain Template code on line 31 in Obviously, you could modify this to ingest data from whichever source you'd like.


  1. It's important we execute the steps above in a virtual environment to prevent unexpected issues with pip installations.
  2. This tutorial was tested using Python 3.10 if you experience issues with the aiohttp package double check your Python version.
  3. If you're getting a RuntimeError: no validator found for <class 'rag_conversation.chain.ChatHistory'>, see 'arbitrary_types_allowed' in Config downgrade your pydantic package to version 1.10.8
Thanks Adam for pointing out and contributing to solving the issues above.

Final Thoughts

Awesome! We've done a lot in this post and LangChain Templates were just released a few hours ago. It's really interesting to see all the great features coming to the framework in such a short time it's hard to keep up. Great work by the LangChain team ๐Ÿ‘๐Ÿ‘๐Ÿ‘

I'd love to know how you plan on integrating these templates in your application so please share your comments, feedback, and questions in the comments below.

Finally, here's the official GitHub repo that contains all of the launched LangChain Templates:

langchain/templates at master ยท langchain-ai/langchain
โšก Building applications with LLMs through composability โšก - langchain-ai/langchain


Hope you found this helpful in your coding journey. If you have, I'd really like to connect with you to learn more about your business and what you're building. So please follow me on X and make sure you subscribe for free to the blog so you're always one step ahead of everyone else.

Cheers ๐Ÿป