Give Your LLM a Memory: RAG with File Upload and Web Crawl

What This Video Covers

Two ways to give an agent knowledge, with no code: upload a document, or crawl a website. The video creates a legal source from an uploaded PDF and a technical source by crawling the TheYgent docs (capped at 20 pages, with optional JavaScript rendering for sites that need a browser). Both are embedded by a local Nomic embedding model into pgvector in your own Postgres, and the crawl log shows which pages made it in.

Then it builds an agent on the canvas: an LLM with two retrieval tool nodes, each with a description that tells the model when to use it. In the test panel, a general question uses no tool, a question about data retention pulls section 4.3 of the PDF, and a question about importing a website searches the crawled docs — each tool call visible in the run trace. Publishing it makes the agent available over its endpoint.

The LLM in this demo is a hosted model; any model with tool calling, local or hosted, works the same way.

Try It Yourself

• RAG sources
• The RAG node
• Installing an embedding model