Open WebUI and Granite 4.1: local RAG setup

In this Spanish-language tutorial, Hugo Paredes walks through a local RAG setup with Open WebUI and IBM Granite 4.1, using 10 legal contracts as the document collection. The tutorial starts with Docker Desktop and Ollama Desktop already installed on Windows. He selects the 8B model for his RTX 3070 with 8 GB of VRAM and shows a download of about 5 GB.

He raises Ollama's context setting in Open WebUI from 2,048 to 16,384. Paredes frames context configuration as a way to avoid exhausting GPU memory, but the demonstration does not measure VRAM use or establish that this increase reduces it. His hardware choice and reported performance describe this example, rather than requirements or timings for every machine.

In Workspace's Knowledge section, he creates a collection, uploads the 10 PDFs and attaches that knowledge to a new chat. The prompt asks for a comparison table with each filename, contract type, effective date, termination clause and legal jurisdiction, followed by a judgment about the riskiest jurisdiction. He reports that the response finishes in roughly 30 seconds; the tutorial does not independently validate the legal assessment.

For viewers who want to run models locally against private documents, the practical focus is the knowledge collection and chat workflow. Paredes says the files stay on local hardware, though the demonstration does not establish a universal security guarantee.