## Metadata - Author: [[arcturus-labs.com|Arcturus Labs]] - Full Title:: Roaming RAG – RAG Without the Vector Database - Category:: #🗞️Articles - URL:: https://arcturus-labs.com/blog/2024/11/21/roaming-rag--rag-without-the-vector-database/ - Finished date:: [[2024-12-10]] ## Highlights > Jeremey Howard proposed llms.txt ([here](https://llmstxt.org/)) as a machine-readable document that is associated with a website. Similar to /sitemap.xml and /robots.txt, llms.txt is intended to be read by machines, and it serves as a structured guide to help large language models quickly understand the key information about a website or project. ([View Highlight](https://read.readwise.io/read/01jerdtjnwkj7jsc49qve4qdsv)) ## New highlights added [[2024-12-14]] > The most important thing for our purposes is that llms.txt is formatted as a well-structured markdown document – something that is both easy for LLMs to consume and easy for *me* to use in a demo. ([View Highlight](https://read.readwise.io/read/01jf2np7n7hcpb9yajkwxw2p23)) > If the model calls `expand_section` then the corresponding section of the document is retrieved and presented to the model in another artifact. (The artifact is also presented to the user as a new tab in the artifact pane. Using artifacts isn't required for Roaming RAG, but it was a convenient choice coming from my last blog post, and it seemed to work well here too.) ([View Highlight](https://read.readwise.io/read/01jf2nsfpj2dd4ed84nmeagnmf))