I followed this video tutorial by Savana AI:

I'm hosting Flowise on a Render free plan (and connected it to MongoDB for persistent storage). I'm using Google Gemini for embeddings, Pinecone for storing and searching through the embeddings, and also Google Gemini for the text generation. From there, I connected all of the components to a conversational retriever QA chain component that's built into Flowise (let me know if any of that doesn't make sense)

I'm also using make.com to send messages to the Flowise workflow, and saving my data on Airtable

However, the airtable step is confusing me. My data is not question-answer pairs, its just chunks of text that I want the RAG system to search through for loading into the model's context window

I realized that I'm not really sure what flowise is doing with the data I feed it from airtable. Does it take each row and encode it as a vector? Or encode each question and answer separately?

I also don't fully understand the conversation - retrieval QA chain from Flowise. how does it work? And does Flowise know only feed the answer from the document it found? Or both the question-answer pair

Hi, I checked out the knowledge base tool and it works great! However, I wanted to ask about how it works. Does the query tool load the whole knowledge base into the model's context window? If so, how can I use the divided dataset to load only the most similar chunk (similar to how a RAG does this)?