RAG Pipeline

purpose to create and manage a retrieval augmented generation (rag) pipeline that connects data sources, embeddings, and vector stores this node enables intelligent context retrieval for ai agents and workflows — improving accuracy and relevance in language model responses problem statement large language models (llms) have limited context windows and cannot access private or proprietary data by default manually integrating document parsing, embeddings, and vector databases requires complex setup the rag system node provides a no code interface to build, manage, and query your rag pipeline — connecting data ingestion, vectorization, and retrieval seamlessly requirements must support multiple data sources (files, apis, whatsapp, voice, etc ) should handle chunking and preprocessing of text automatically must integrate with major embedding providers (e g , openai, hugging face) should connect to a vector database (e g , supabase, pinecone, qdrant) must store and return retrieval results as workflow variables should support both plain text and json output formats use cases enable chatbots to reference custom documents, files, or transcripts build knowledge grounded question answering systems implement context aware assistants for business or internal tools connect workflows to ai memory systems combine with llm or openai nodes for advanced rag based reasoning outcome creates a fully configured data retrieval pipeline that indexes, stores, and retrieves relevant information for ai powered nodes and workflows how to use 1\ drag & drop add the rag system node into your workflow canvas this node acts as the data foundation for retrieval augmented operations data storage pipeline the rag pipeline consists of four configurable stages , visible in the main dashboard 1 data source 2 processing 3 embeddings 4 vector store 5 advanced variable input use any variable as a rag data source like file upload, add variable name in the variable name field, specify which variable to use as your input source data sources add and configure the data inputs that will feed into your rag system files pdf, txt, docx, csv, etc once configured, the connected data source will automatically send its content to the next rag pipeline stages processing → chunking and text segmentation embeddings → vector generation vector store → data storage and retrieval processing chunk and preprocess your input text before embedding options may include chunk size (e g , 500–1000 tokens) overlap (for contextual continuity) text normalization (lowercase, punctuation cleanup) embeddings transform processed chunks into vector representations using your selected provider you can configure model name batch size embedding dimension rag response generation how to use 1\ access response generation panel open the rag system configuration → response generation tab this section defines how your ai model interacts with stored rag data 2\ connect a rag system ensure you have an existing rag system configured under data store select your active rag pipeline to use as the data retrieval source 3\ select model provider choose the ai provider and model for response generation 4\ configure prompt inputs the prompt defines how the model uses retrieved data to generate responses 5\ retrieval options specify how many context chunks to fetch from the vector store 6\ output variable name set the variable to store the ai generated answer default ragresponse 7\ output format choose how the response should be structured plain text — for direct user display json — for structured data workflows 8\ advanced settings you can adjust model parameters for fine tuning response behavior