OneRAG is an open-source RAG (Retrieval-Augmented Generation) project developed by a non-developer product manager to streamline the creation and customization of RAG chatbots. The project addresses the inefficiency of repeatedly modifying code for varied client requirements concerning RAG pipelines and vector database integrations, a challenge encountered during the development of numerous RAG chatbot products on a freelancing platform, which generated approximately 14 million KRW in 5 months.

The core methodology of OneRAG is its highly modular and configurable architecture. It allows for the dynamic interchangeability of various RAG pipeline components by simply modifying a single line within a configuration file. This is achieved through an abstraction layer that defines interfaces for different component types, enabling users to switch between specific implementations without altering the underlying code logic.

Supported components include:

• Vector Databases: Weaviate, Chroma, Pinecone, Qdrant, pgvector, MongoDB.
• Large Language Models (LLMs): Gemini, OpenAI, Claude, OpenRouter.
• Rerankers: Jina, Cohere, Google, OpenAI, Local (implying local models or custom reranking logic).
• Caching Mechanisms: Memory, Redis, Semantic.
• Additional Features: GraphRAG capabilities, PII (Personally Identifiable Information) Masking, and Agent functionalities.

Technically, OneRAG is built on FastAPI, a modern Python web framework, and includes Docker Compose support for simplified deployment and environment setup. It offers native support for Korean NLP, reflecting the developer's background in Korean AI services. The project is released under the MIT License, encouraging open collaboration and use. A quickstart guide is provided, involving git clone https://github.com/notadev-iamaura/OneRAG followed by cd OneRAG && make quickstart. The developer is actively seeking community feedback on its architecture and suggestions for incorporating additional components.