OpenRAG is a package to build agentic Retrieval Augmented Generation applications that comes with all necessary components out of the box. It’s a new open-source initiative led by IBM leveraging other popular open-source projects.
OpenRAG comes with three main components:
- Docling: Text extraction from unstructured documents
- Langflow: Graphical tool to build agentic flows
- OpenSearch: Fork of Elasticsearch for semantic searches
The three open-source projects are well integrated in OpenRAG. Within minutes you can build your first RAG applications locally. Models can be integrated from various providers including local models via Ollama. Knowledge can be uploaded and accessed from AWS, Google and Microsoft.
Introduction
OpenRAG was announced and introduced recently in the OpenRAG Summit.
Models
Models can be accessed from OpenAI, Antrophic, watsonx.ai and Ollama.
Flows
OpenRAG utilizes Langflow to run the ingestion pipelines and the inferencing. There are default implementations, but as a developer you can also modify everything.
To access OpenRAG programmatically, the Langflow APIs can be leveraged, or you can embed the functionality as MCP tools.
Data
Data can be uploaded and accessed from other popular sources. Filters can be set and advanced queries can be run via OpenSearch features.
Developers can also modify the ingestion parameters or the full Langflow ingestion pipeline.
Next Steps
To find out more, check out the following resources: