OpenRAG 0.4 is a self-hosted Retrieval-Augmented Generation (RAG) package developed by Langflow and IBM. According to Discovery materials, it was released on March 31, 2026. It combines Docling, which structures documents into a readable format; OpenSearch, which handles search indexing and vector/semantic search; and Langflow, which connects search results to generative AI and agent flows, into a single configuration. Similar to how a person would classify and index documents in a filing cabinet and then retrieve and deliver the necessary documents to the responsible person, OpenRAG is a tool that integrates the RAG processing steps, from unstructured document collection to search and response workflows. In a typical RAG implementation, you would need to select and connect a PDF parser, document splitter, embedding processor, vector search engine, prompt chain, and operational interface. This process can lead to issues such as the loss of table and document hierarchy, or the failure to properly transmit metadata created during the collection phase to the search phase. The key differentiator of OpenRAG is that it pre-combines Docling-based document structuring, OpenSearch-based search, and Langflow-based orchestration. Instead of providing only specific algorithms as a single library, it takes an approach of handling everything as a single self-hosted package, from the point where documents enter to the point where search results are delivered to the agent. Life science researchers can collect data with different structures, such as journal PDFs, experimental reports, Standard Operating Procedures (SOPs), and analysis forms, and then build a document-based question-answering environment. For example, they can extract the title, body, and tables of a journal using Docling, index the original text and metadata in OpenSearch, and then design a RAG flow in Langflow to deliver the search results to a language model. Researchers can ask questions about specific experimental conditions or analytical methods and review the responses, which include the retrieved source documents, to assist with literature reviews and internal knowledge exploration. However, the supported file formats, splitting parameters, embedding models, and citation output methods require verification in the official documentation. For pharmaceutical and biotech organizations that want to store quality documents and research records in their own environment while operating search and agent workflows, OpenRAG can be a candidate solution. By delivering OpenSearch search results to subsequent review stages or separate workflows in Langflow, it is possible to design a configuration that continuously processes document exploration, comparison, summarization, and review requests. However, data security features such as access control, audit logs, encryption, and regulatory compliance cannot be confirmed based on the provided Discovery information alone, so the official documentation and deployment configuration must be verified separately before applying it to clinical and regulatory tasks.

OpenRAG 0.4 is a self-hosted Retrieval-Augmented Generation (RAG) package, reportedly developed by Langflow and IBM. According to Discovery materials, it was released on March 31, 2026, and combines Docling, which structures documents into a readable format; OpenSearch, which handles search indexing and vector/semantic search; and Langflow, which connects search results to generative AI and agent workflows, into a single configuration. It allows users to classify and index materials in a filing system and then retrieve the necessary documents.

Literature-Based Evidence Retrieval: Extract the text and table structures from research paper PDFs using Docling, index them in OpenSearch, and then pass them to a Langflow RAG flow to review answers and supporting documents related to experimental conditions or analytical methods. Refer to the official documentation for details on chunk size, number of search results, and accuracy.