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.

Demo(s)

Launching the server locally, running the client, and pipeline observabiilty application:

!! Note - The server has been removed from this repo - instead we now recommend using SciPhi Cloud to pair with the R2R framework for observability and optimization.

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-...
export LOCAL_DB_PATH=local.sqlite

# OR do `vim .env.example && cp .env.example .env`
# INCLUDE secrets and modify config.json
# if using cloud providers (e.g. pgvector, supabase, ...)

Run the server with Docker:

docker pull emrgntcmplxty/r2r:latest

# Place your secrets in `.env` before deploying
docker run -d --name r2r_container -p 8000:8000 --env-file .env r2r

Links

Join the Discord server

Read the R2R Docs

Basic Examples

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

basic app: This example runs the backend server, which includes the ingestion, embedding, and RAG pipelines served via FastAPI.
```
# If using a venv, replace `uvicorn` with `venv_path/bin/uvicorn`
uvicorn r2r.examples.basic.app:app
```
basic client: This example should be run after starting the server. 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
```

academy: A more sophisticated demo demonstrating how to build a more novel pipeline which involves synthetic queries

# Launch the `academy` example application
# If using a venv, replace `uvicorn` with `venv_path/bin/uvicorn`
uvicorn r2r.examples.academy.app:app

# Ask a question
python -m r2r.examples.academy.run_client search "What are the key themes of Meditations?"

end-to-end: An example showing how to combine a complete web application with the basic RAG pipeline above.
intelligence: A cloud platform which can be used to deploy R2R pipelines powered by SciPhi

Full Install:

Follow these steps to ensure a smooth setup:

Install Poetry:
- Before installing the project, make sure you have Poetry on your system. If not, visit the official Poetry website for installation instructions.
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 -E parsing -E eval

Execute with poetry run:

python -m r2r.examples.pdf_chat.run_client ingest

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 Standardization: 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.

haranrk/R2R