An AI that can show you where it got that, not just tell you.
Lexi searches a curated library and quotes it with citations, so you can check her instead of trusting her.
Retrieval, not recall
Most models answer from memory. Training compresses a text into weights, and what comes back out is a paraphrase with no way to tell a real quotation from a confident reconstruction. That is where fabricated citations come from, and it is worst on exactly the material people care most about getting right.
Lexi searches the library and reads the passage before she answers. When she quotes, the words are the words, and she names the source.
What is in it
A curated corpus spanning theology, history, philosophy, politics, traditionalism and comparative religion. It is a working library rather than a scrape: every source is deliberately chosen, and she can list what is available in a subject before she searches it.
Material the big models will not engage with
Ask a mainstream assistant about a contested historical or theological question and you generally get a redirect. Not because the sources do not exist, but because engaging is a product risk for them.
Lexi reads the material and argues from it. You can disagree with her reading, which is the point, and you can go to the source she cited and check.
And the live web when the question is new
The library is for depth. For anything recent, Lexi searches the web and tells you what she found, so a question about this week does not fall into the gap between training data and today.