Maya: Facts You Can Check, a Story That Coheres

News-feed RAG where every claim is traceable — and scattered events stitch into a developing story.

A news feed is a firehose. A thousand fragments a day arrive orphaned from the ones before them: a headline with nothing under it, a claim you cannot easily check, a moment with no memory of the moment that led to it. You scroll, and almost none of it connects.

Maya — the assistant inside Knobly Cream — is built to fix the two things that make a feed hard to trust and hard to follow. It does not answer from a model's memory. It answers from the feed itself: ask it something and it retrieves the actual cards — the posts, the news items — that bear on the question, and writes from those. That is the RAG part, retrieval-augmented generation, and it changes what an answer even is.

The first payoff is verifiability. Every sentence Maya gives you is anchored to the source it came from. You can follow the thread back to the card, to who said it and when it entered the feed, and judge it for yourself. A claim with a receipt is a different animal from a claim floating free — and a feed where every fact carries its provenance is a feed you can actually rely on.

But the larger shift — the one that matters more — is the stitching. A single news item is a snapshot: maybe true, but flat, and frozen. The same event throws off a dozen snapshots over days — a first report, a correction, a reaction, a consequence. Maya recognizes that those scattered cards are about the same unfolding thing and threads them into one developing story: a narrative with a spine and a clock, that grows as new facts land instead of resetting with every fresh headline.

So you stop reading the news as confetti and start reading it as a story in progress — here is what is known, here is exactly where each piece came from, here is how it got here, and here is what changed since yesterday. Facts you can check, woven into a thread you can follow.

None of this is free. Pulling the right cards out of a live feed, deciding which fragments belong to the same story, keeping it all current — that is retrieval and embeddings at speed, and it runs on the kind of hardware the rest of this blog is about: GPUs a few feet away, quietly doing the stitching.