Generative UI with LangChain Python 🦜🔗

Overview

This application aims to provide a template for building generative UI applications with LangChain Python. It comes pre-built with a few UI features which you can use to play about with gen ui. The UI components are built using Shadcn.

Getting Started

Installation

First, clone the repository and install dependencies:

git clone https://github.com/bracesproul/gen-ui-python.git

cd gen-ui-python

Install dependencies in the frontend and backend directories:

cd ./frontend

yarn install

cd ../backend

poetry install

Secrets

Next, if you plan on using the existing pre-built UI components, you'll need to set a few environment variables:

Copy the .env.example file to .env inside the backend directory.

LangSmith keys are optional, but highly recommended if you plan on developing this application further.

The OPENAI_API_KEY is required. Get your OpenAI API key from the OpenAI dashboard.

Create a new GitHub PAT (Personal Access Token) with the repo scope.

Create a free Geocode account.

# ------------------LangSmith tracing------------------
LANGCHAIN_API_KEY=...
LANGCHAIN_CALLBACKS_BACKGROUND=true
LANGCHAIN_TRACING_V2=true
# -----------------------------------------------------

GITHUB_TOKEN=...
OPENAI_API_KEY=...
GEOCODE_API_KEY=...

Running the Application

cd ./frontend

yarn dev

This will start a development server on http://localhost:3000.

Then, in a new terminal window:

cd ../backend

poetry run start

Go further

If you're interested in ways to take this demo application further, I'd consider the following:

Generating entire React components to be rendered, instead of relying on pre-built components.
Using the LLM to build custom components using a UI library like Shadcn.
Multi-tool and component usage.
Update the LangGraph agent to call multiple tools, and appending multiple different UI components to the client rendered UI.
Generative UI outside of the chatbot window: Have the UI dynamically render in different areas on the screen. E.g a dashboard, where the components are dynamically rendered based on the LLMs output.