[d4rkp4ttern]

Web search is not a capability of a “bare” LLM, but in an LLM-based system it can be done by giving the LLM access to a “web search tool”, I.e essentially you instruct it to output a specific structured text (typically json but doesn’t have to be) indicating its “intent” to search, and your wrapper intercepts/detects this structured response, does the actual search and returns the results (e.g snippets from top k results) into the context of the LLM amd have it use these to respond to your question.

A similar thing can be done with external documents - your wrapper retrieves docs/fragments relevant to the query, puts them in the context and lets the LLM use them to answer the query. This is called Retrieval Augmented Generation (RAG).

The above is a highly simplified description. In the Langroid library (a multi-agent framework from ex-CMU/UW-Madison researchers) we have these and more. For example here’s a script that combines web search and RAG:

https://github.com/langroid/langroid/blob/main/examples/docq...