LLMs are great but don’t have access to the internet inherently. They also don’t know what you know.

Our knowledge map is the answer (which can be used alongside our search and Deep Dive modes).

What is a Knowledge Map?

Your knowledge map is built from any reference material that you want to feed into your chats, in an intelligent, on-demand way. This can be anything from a PDF to a website to just some code or text you paste in that you want to be referenced now and then.

How do I use it?

You first click “Knowledge Map” in the bottom of the left sidebar.

This will open up a form where you can upload anything you want to reference in chats.

It looks a little bit like this:

Now, you can upload your PDFs, CSVs, or put in a URL to scrape a website (useful for coding reference docs, for example).

You can also paste in text, or code, or … whatever else you want to reference.

NOTE: You also need to set “Use Knowledge Map in Replies” in the settings (gear icon) to enable this. You can toggle it on/off as you wish, but it’s off by default, since it uses more credits than a normal prompt (because it’s pulling all the most relevant context from your real documents, NOT just a summary).

Secret hint: play with turning it on and off for different queries, you’ll find sometimes you want those and sometimes you would rather have a faster result without. I tweak the setttings constantly - that’s why we leave it just a click or two to change.

How does it work?

Technically, our Knowledge Map is a sophisticated form of retrieval-augmented generation (RAG).

Ours is unique, in that we mix a host of different search techniques above and beyond the basic vector-based search commonly used in RAG systems.

This ensures we get class-leading accuracy, so you can stuff in everything you want, and we’ll pull only the best, most relevant info out at any given time.

How we do that technically is by scoring relevance based on several simultaneous signals - in our case, a knowledge graph, full text search, and a souped-up version of vector-based search (not just the classic, but improved with techniques from the bleeding edge of AI research).

How we construct each of those pieces is a competitive advantage, but now you know why your custom Knowledge Map is not “just” RAG or extracted details, like others.