llama_deploy/examples/python_fullstack

Python Fullstack Example

This example demonstrates a fullstack application using llama-deploy to create and manage a RAG (Retrieval-Augmented Generation) and an Agentic workflow.

Overview

The application consists of two main workflows:

1. RAG Workflow: A basic retrieval-augmented generation system.
2. Agentic Workflow: An advanced workflow that incorporates the RAG system and adds agentic capabilities.

These workflows are deployed as separate services using llama-deploy, allowing for flexible and scalable deployment options.

Then, a simple frontend is built using reflex to allow you to chat with the deployed RAG workflow and the agentic workflow.

Project Structure

Let's walk through the important files and folders:

• frontend/: A simple frontend built using reflex to allow you to chat with the deployed RAG workflow and the agentic workflow.
  • frontend/frontend/frontend.py: The reflex app definition. Builds a basic chat UI.
  • frontend/frontend/state.py: The state management for the frontend. This is where we actually connect to the llama-deploy api to chat with the workflows.
  • frontend/frontend/style.py: The style management for the frontend. This is where we define the style of the chat UI.
• workflows/: The workflows themselves, including the RAG workflow and the agentic workflow.
  • workflows/agent_workflow.py: The agentic workflow that uses the RAG workflow.
  • workflows/rag_workflow.py: The RAG workflow. This includes indexing with a qdrant vector store, retrieval, reranking with RankGPT, and a response synthesis step.
  • workflows/deploy.py: The entrypoint for the dockerfile, which is used to deploy the workflows. The DEPLOY_SETTINGS_NAME environment variable is used to determine which workflow to deploy.

To orchestrate the core components of llama_deploy all you have to do is to include in your docker-compose.py one of the pre-built compose files from this repository, like this:

include:
  - path: ../../docker/docker-compose-redis.yml

Dependencies

The project relies on several key libraries:

• llama-deploy: For service deployment and management.
• llama-index: For building and running the workflows.
• Various llama-index extensions for specific functionalities (e.g., RankGPT, QdrantVectorStore).

Usage

1. Ensure you have docker installed and running.
2. Export your OpenAI key in your terminal (i.e. export OPENAI_API_KEY="...")
3. Run docker compose up from the root of the project (in the examples/python_fullstack directory) to start the frontend and backend services.
4. Open your browser and navigate to http://localhost:3000 to access the chat interface and chat with the deployed RAG workflow and agentic workflow.

Service Details

• Control Plane:

  • Port: 8000

• RAG Workflow Service:

  • Port: 8002
  • Service Name: "rag_workflow"

• Agentic Workflow Service:

  • Port: 8003
  • Service Name: "agentic_workflow"

• Frontend:

  • Port: 3000

• Qdrant:

  • Port: 6333

The message queue service can be different, depending on the backend of choice. In this example, Redis is used, so you would also see this service running:

• Redis

  • Port: 6379

Extensibility

This example serves as a foundation for building more complex applications. You can extend the workflows, add new services, or integrate with other components of your system as needed.