Why "Similar Text" Isn't the Same as an Answer...
TL;DR: Opt your attachments into Lex's knowledge graph and Lex can connect facts across multiple documents to answer questions that keyword or semantic search alone can't solve.
The problem: search that matches words, not meaning
Keyword search and semantic search both do the same basic thing - they find text that looks like your question. Neither one actually reasons about what that text means.
You search for a symptom and get back documents that mention the same words, but none of them explain the underlying cause
The documents "match," but they don't answer the question
A familiar example
Say a customer's Outlook or CRM performance is slow. The real answer often isn't sitting in one document - it's scattered across several documents that never even share the same keywords. One might cover network configuration, another might cover a specific application setting, and another might cover a hardware limitation.
Relevance isn't about ranking the paragraphs that look the most similar to your question. It's about connecting the facts that explain it.
What a knowledge graph does differently
When an attachment is added to Lex's knowledge graph, Lex breaks it into chunks, summarizes the concepts inside it, and builds typed relationships between those concepts - entity, relationship, entity.
The result is a reasoning trajectory. Instead of stopping at "here's some similar text," Lex can traverse from your question, through related concepts, to an actual answer.
đ Knowledge graphs aren't a new or secret technology - what matters is that Lex applies one directly to your MSP documentation, so the concepts inside your attachments become something Lex can reason across.
What this means in practice
Opting an attachment into the knowledge graph takes one click - once it's in, it's no longer just a searchable file, it's part of Lex's understanding of your documentation
Lex uses knowledge graph search automatically as one of its tools when answering questions
If Lex doesn't reach for it on the first prompt, you can ask it directly to search the knowledge graph
đĄ For step-by-step instructions on opting attachments into the knowledge graph, see [link to how-to article].
Where this is headed
The goal isn't more complex search - it's faster, more precise answers over your MSP documentation. As more of your attachments become part of the knowledge graph, Lex's ability to connect the dots across your environment only gets stronger.