R2R: Production-ready RAG systems.

A semi-opinionated RAG framework.

R2R was conceived to bridge the gap between experimental RAG models and robust, production-ready systems. Our semi-opinionated framework cuts through the complexity, offering a straightforward path to deploy, adapt, and maintain RAG pipelines in production. We prioritize simplicity and practicality, aiming to set a new industry benchmark for ease of use and effectiveness.

Quick Install:

Install R2R directly using pip:

# use the `'r2r[all]'` to download all required deps
pip install 'r2r[parsing,eval]'

# setup env
export OPENAI_API_KEY=sk-...

Links

Join the Discord server

Read our Docs

Basic Examples

The project includes several basic examples that demonstrate application deployment and interaction:

1. app.py: This example runs the main application, which includes the ingestion, embedding, and RAG pipelines served via FastAPI.

   uvicorn r2r.examples.basic.app:app

2. run_client.py: This example should be run after starting the main application. It demonstrates uploading text entries as well as a PDF to the local server with the python client. Further, it shows document and user-level vector management with built-in features.

   python -m r2r.examples.basic.run_client

3. run_pdf_chat.py: An example demonstrating upload and chat with a more realistic pdf.

   # Ingest pdf
   python -m r2r.examples.pdf_chat.run_demo ingest
   
   # Ask a question
   python -m r2r.examples.pdf_chat.run_demo search "What are the key themes of Meditations?"

4. web: A web application which is meant to accompany the framework to provide visual intelligence.

   cd $workdir/web && pnpm install
   
   # Serve the web app
   pnpm dev

Demo

slim_demo.mp4

Full Install:

Follow these steps to ensure a smooth setup:

1. Install Poetry:

   • Before installing the project, make sure you have Poetry on your system. If not, visit the official Poetry website for installation instructions.

2. Clone and Install Dependencies:

• Clone the project repository and navigate to the project directory:

  git clone git@github.com:SciPhi-AI/r2r.git
  cd r2r

• Copy the .env.example file to .env. This file is in the main project folder:

  cp .env.example .env
  
  # Add secrets, `OPENAI_API_KEY` at a minimum
  vim .env

• Install the project dependencies with Poetry:

  # See pyproject.toml for available extras
  # use "all" to include every optional dependency
  poetry install --extras "parsing"

• Execute with poetry run:

  python -m r2r.examples.pdf_chat.run_demo ingest

3. Configure Environment Variables:
   • You need to set up cloud provider secrets in your .env. At a minimum, you will need an OpenAI key.
   • The framework currently supports PostgreSQL (locally), pgvector and Qdrant with plans to extend coverage.

Key Features

• 🚀 Rapid Deployment: Facilitates a smooth setup and development of production-ready RAG systems.
• ⚖️ Flexible Standarization: Ingestion, Embedding, and RAG with proper Observability.
• 🧩 Easy to modify: Provides a structure that can be extended to deploy your own custom pipelines.
• 📦 Versioning: Ensures your work remains reproducible and traceable through version control.
• 🔌 Extensibility: Enables a quick and robust integration with various VectorDBs, LLMs and Embeddings Models.
• 🤖 OSS Driven: Built for and by the OSS community, to help startups and enterprises to quickly build with RAG.
• 📝 Deployment Support: Available to help you build and deploy your RAG systems end-to-end.

Core Abstractions

The framework primarily revolves around three core abstractions:

• The Ingestion Pipeline: Facilitates the preparation of embeddable 'Documents' from various data formats (json, txt, pdf, html, etc.). The abstraction can be found in ingestion.py.

• The Embedding Pipeline: Manages the transformation of text into stored vector embeddings, interacting with embedding and vector database providers through a series of steps (e.g., extract_text, transform_text, chunk_text, embed_chunks, etc.). The abstraction can be found in embedding.py.

• The RAG Pipeline: Works similarly to the embedding pipeline but incorporates an LLM provider to produce text completions. The abstraction can be found in rag.py.

• The Eval Pipeline: Samples some subset of rag_completion calls for evaluation. Currently DeepEval is supported. The abstraction can be found in eval.py.

Each pipeline incorporates a logging database for operation tracking and observability.