Europe 2026
OpenRAG: An open-source stack for RAG — Phil Nash
Overview
This talk introduces OpenRAG, an open-source stack designed to simplify the creation of powerful and customizable Retrieval-Augmented Generation (RAG) systems. It addresses the complexity often encountered in RAG pipelines, from document ingestion and chunking to embedding and search, by integrating existing open-source tools into a cohesive framework. OpenRAG aims to provide a robust baseline that developers can easily extend to meet specific project requirements.
Who should watch
- AI engineers and builders looking for a flexible RAG framework.
- Developers facing challenges with document processing, especially PDFs.
- Teams needing to integrate various embedding models and search strategies.
- Those interested in agentic retrieval for more dynamic search capabilities.
- Individuals seeking an offline-capable RAG solution.
Key takeaways
- RAG remains relevant despite advancements in large context windows, as it offers a cost-effective and efficient way to manage business data.
- OpenRAG is built upon three core open-source projects: Docling for document processing, OpenSearch for indexing and search, and LangFlow for orchestration and agent creation.
- Docling offers flexible pipelines for ingesting diverse document types, including advanced options for PDFs using small models or a vision-language model, and supports hierarchical chunking.
- OpenRAG supports multiple embedding providers and can be run entirely offline, making it suitable for air-gapped environments.
- OpenSearch is utilized for hybrid vector and keyword search, with OpenRAG defaulting to the JVector KNN plugin for efficient vector indexing.
- LangFlow provides a visual interface for building RAG pipelines and agentic retrieval, allowing agents to perform multiple searches as needed.
- Agentic retrieval empowers models to decide which searches to perform and how to use the results, moving beyond traditional top-K retrieval.
- OpenRAG is highly customizable, allowing users to configure model providers, agent instructions, chunking strategies, and even integrate external document stores via cloud connectors.
Notable quotes
*Rag is quite complex. In fact, everyone's documents are different, every system will have different users, different questions, different interaction patterns and different expectations.*
*OpenRag is an opinionated, but agentic, and open source stack for RAG.*
Unofficial community note. Prefer the recording for nuance.