Vision RAG & Image Embeddings with Cohere Command-A and Embed v4

This workflow leverages Cohere's new multimodal models (Command-A-vision and Embed v4) to build a Vision RAG system for efficient document scan retrieval and analysis. It downloads page extracts from reports with graphs/charts, converts them to base64, generates embeddings, and stores them with original URLs in a Qdrant vector database using the community node.

The agent splits into a regular chat AI for conversation and a specialized Q&A branch powered by Command-A-vision. On relevant queries, it performs vector search to fetch matching image URLs, feeds them to the vision model with the user query, and responds seamlessly. Both agents share memory for a unified chat experience.

Benefits include reduced context costs (no need to load all images), querying graphical content like charts/tables, and scalable multimodal RAG. Ideal for tech reports, invoices, or any image-based docs.

To use: Set up Cohere API keys, replace sample pages with your documents (split/convert if needed), configure Qdrant, and customize prompts. Perfect for knowledge bases, research assistants, or automated document Q&A.