Langflow just launched OpenRAG, a 100% open source RAG platform that packages everything you need to build a semantic search and document chat system into a single command: ingestion, indexing, search, and visual flows.
One command: uvx openrag. No ad-hoc integrations. No vendor lock-in.
1. Quick Recap: What Is RAG?
RAG (Retrieval-Augmented Generation) is the dominant architecture for building applications that let LLMs answer questions about specific information that wasn't in their training data: internal documents, knowledge bases, contracts, technical manuals, etc.
The basic RAG flow
- 1. Ingestion: Documents are processed, split into chunks, and converted into embeddings (numerical vectors).
- 2. Indexing: Those embeddings are stored in a vector database for efficient search.
- 3. Retrieval: When the user asks a question, the most relevant chunks are found by semantic similarity.
- 4. Generation: The LLM receives the question plus the retrieved context and generates a grounded response.
The problem until now was that assembling a complete RAG pipeline required manually integrating an orchestration framework, an embeddings model, a vector database, and a user interface. OpenRAG solves all of that at once.
2. What Is OpenRAG?
OpenRAG is a ready-to-use open source distribution built by the Langflow team, packaging three well-established tools into a cohesive system:
Langflow
Orchestration engine with a visual flow editor. Connect models, tools, and data sources without code.
Docling
IBM's document processing pipeline. Extracts structured text from PDFs, Word, Excel, images and more with OCR.
OpenSearch
Open source search engine (Elasticsearch fork) that acts as a vector database for semantic search.
What makes OpenRAG special
It's not simply a wrapper around these tools. It's a pre-configured integration where each component is connected, tested, and production-ready. You don't need to configure anything manually.
3. What OpenRAG Includes
Document Ingestion Pipeline
You upload a document (PDF, Word, image, etc.) and the system automatically processes it with Docling: extracts the text, splits it into semantically coherent chunks, generates the embeddings, and indexes them in OpenSearch. All in a continuous flow.
Chat with Documents
Native chat interface for asking questions about indexed content. The system retrieves the relevant fragments and passes them to the LLM to generate answers with real context, not hallucinations.
Native Semantic Search
OpenSearch integrated as a vector store. Supports semantic similarity search (embeddings) and traditional lexical search (BM25), with the option to combine both (hybrid search).
Visual Flows with Langflow
The Langflow visual editor is included, allowing you to modify or extend RAG flows without touching code. You can add reranking steps, filters, conversational memory, or connect external APIs by dragging components.
Docker Support
If you prefer to run it in containers, it has official Docker support. Ideal for deploying on any server or integrating into existing CI/CD pipelines.
4. How to Run It
OpenRAG's value proposition is installation simplicity. There's no need to configure services separately or edit configuration files to get started.
Option 1 — uvx (recommended, no prior installation)
uvx openrag
Option 2 — pip
pip install openrag
openrag
Option 3 — Docker
git clone https://github.com/langflow-ai/openrag
cd openrag
docker compose up
What is uvx?
uvx is the tool runner for uv, the new ultra-fast Python package manager (written in Rust). It lets you run Python packages directly without installing them globally, similar to npx in the Node.js ecosystem.
5. The Technical Stack in Detail
Langflow as Orchestrator
Langflow is an LLM orchestration framework with a visual interface. In OpenRAG it acts as the system's brain: it defines how information flows from ingestion to the final response. Its flows are exportable, versionable, and deployable JSON files.
This means you can customize the RAG pipeline without touching infrastructure code. Changing the embeddings model, adding a reranking step with Cohere, or connecting an additional data source takes minutes from the visual editor.
Docling for Document Processing
Docling is IBM's open source project for document conversion. Unlike simpler solutions, it preserves document structure when processing: tables, lists, headings, and hierarchies are maintained, which significantly improves chunk quality and therefore retrieval precision.
Formats supported by Docling:
PDF · DOCX · XLSX · PPTX · HTML · Markdown · Images (with OCR)
OpenSearch as Vector Store
OpenSearch is the open source fork of Elasticsearch maintained by Amazon and the community. It includes native vector search support (k-NN) since version 2.0, making it a solid alternative to solutions like Pinecone, Weaviate, or Qdrant, but fully self-hostable.
The advantage of using OpenSearch over a pure vector database is that it also supports lexical search (BM25), enabling hybrid search without additional components.
6. Ideal Use Cases
Internal Knowledge Base
Index technical documentation, process manuals, company policies, or internal wikis so the team can query them in natural language.
Contract and Legal Document Analysis
Load contracts, agreements, or regulations to ask specific questions without reading them in full. "What is the termination clause in contract X?"
Research and Paper Analysis
Index collections of scientific papers or technical reports to accelerate literature reviews and comparative analysis.
Technical Support with Documentation
Build a support assistant that answers questions about products or services based on official documentation, reducing the load on the support team.
7. OpenRAG vs. Building Your Own RAG Stack
The obvious question is: when should you use OpenRAG and when does it make sense to build your own pipeline?
✅ Use OpenRAG if...
- • You need a working RAG quickly, without weeks of setup
- • You want something fully self-hostable with no vendor lock-in
- • Your team is not expert in ML infrastructure
- • Langflow's visual flows fit your use case
- • You want a solid foundation to iterate on
⚠️ Consider a custom stack if...
- • You need full control over every component of the pipeline
- • You have very specific scale requirements
- • You want to use a vector store other than OpenSearch (Qdrant, Weaviate, etc.)
- • Your pipeline has very complex business logic that doesn't fit into visual flows
- • You already have consolidated search infrastructure
My perspective
OpenRAG is excellent as a starting point or for projects where RAG is an enabling component, not the main product. For projects where the RAG pipeline is the core of the business with very specific performance, scale, or customization requirements, you'll probably need more control than a pre-packaged solution can give you. But even in that case, OpenRAG is a very valuable architecture reference.
Conclusion
OpenRAG is the clearest signal yet that the RAG ecosystem is maturing. You no longer need to be an infrastructure expert to have a complete, functional RAG pipeline. Langflow, Docling, and OpenSearch are three solid projects individually; together, well-integrated and available with a single command, they are a proposition worth paying attention to.
If you're evaluating building a semantic search system or document chat, OpenRAG is the best open source starting point available today.
Resources
- • Official site: openr.ag
- • GitHub: github.com/langflow-ai/openrag
- • Langflow: langflow.org
- • Docling: github.com/DS4SD/docling
