@Nate Herk Any relation why usage is high just call the API here?

Thank you, Nate Herk! We could consider enhancing the existing workflow to make it even more unique by adding the following features: 1. Each document contains its own images, tables, graphs, etc. 2. Extract all images from each document and ensure they are carefully assigned to their relevant documents. 3. Embed the extracted images using a CLIP model. Chunking will be critical here—adding additional metadata to image descriptions will help ensure accurate retrieval of relevant images. Store these embeddings in a separate index within the vector database. 4. During retrieval, combine both text embeddings and image embeddings to generate more relevant and accurate results. 5. This approach would make our workflow even more distinctive! I’d be happy to share additional insights on this—although I haven’t explored implementing it in n8n yet.

Thank you James for your detailed response!!! Got more hope to rely on n8n and ready for production. More value information and def this will help for others. Thank you again!!!