TL;DR
- Hi, Moose can now build an entity graph of your site: the companies, products, people, and topics your pages establish, and the relations between them, exportable from chat as CSV or a PDF report.
- The Quick graph reads the schema.org structured data your pages already publish and takes seconds; the Full graph adds an AI pass over your most important pages, on your local model or your own key.
- Site search got rebuilt at passage level, so answers now quote the part of a page that matched your question instead of the first paragraph.
- Every chat now opens with an inventory of what site data the assistant holds, with counts and a last-indexed date, and empty results say whether nothing matched or nothing is indexed.
- A new Search index panel in Settings shows what's indexed and rebuilds it from crawl data you already have.
How assistants read your site
AI assistants don't experience your website the way people do. They extract and retrieve pieces of its content, then infer the companies, products, people, claims, and relationships needed to answer a question. Your site feeds that inference constantly, through headings, structured data, and the way pages reference each other.
Which raises a fair question: what do your pages add up to? Until this week, we couldn't show you. Now we can.
Ask for the graph
Ask the chat for an entity graph of your site and a card opens with two depths.
Quick reads the schema.org markup your pages already publish, the same structured data answer engines read, and turns it into a graph in seconds. Your organization, its products, the people, the places, each entity backed by the pages that establish it, each relation typed: this product belongs to this brand, this person works for this company.
Full goes further. An AI pass reads your most important pages and pulls out the entities the markup never mentioned, the partner named in a case study, the integration buried in paragraph six. That pass runs on your local model or your own OpenRouter key, and it remembers what it has read, so running it again only pays for pages that changed.
The result card exports two CSV files, entities and relations, or a PDF report with evidence URLs included. Hand it to a client, diff it after a content sprint, or just find out which entities your site establishes that you never meant to. All of it comes from your own crawled pages, stored on your own machine.
Search that quotes the middle of the page
Underneath the graph, site search got a rebuild it had been owed for a while.
Until now the index held one embedding per page, built mostly from the title, the headings, and the opening paragraph. Ask about something explained halfway down a long page and the search was squinting at the intro. Pages are now stored as small passages, each searchable on its own, so retrieval returns the sentence-and-a-half that answers your question, wherever it lives on the page.
No re-crawl needed. The new index builds from the crawl data site monitoring already collected, there's a Rebuild index button in Settings with live counts, and the app checks for a stale index at startup and quietly repairs it.
An assistant that knows what it knows
One more change, small in code and large in trust. Every project conversation now opens with the facts about what site data the assistant holds: how many pages are indexed, how many passages, and when they were last updated.
That inventory does quiet work. It means the assistant answers from what it has instead of guessing about what it might have. And when a search comes back empty, it can now tell you which kind of empty: your site doesn't cover this, or your site isn't indexed yet, along with where to fix that. Those are very different sentences to hand a marketer, and a tool that can't tell them apart isn't one you should have to trust.
The honest part
Two edges worth knowing. The Full graph needs a local model or your own OpenRouter key today; managed-plan support for the deeper pass is on the list. And the graph's result card lives in the chat session, so after a restart you regenerate the view, though the graph itself is stored and comes back instantly.
The download is free if you want to see what your site adds up to. I'm curious what yours says about you.
FAQ
What is an entity graph in Hi, Moose?
It's a map of the named things your site establishes, like your company, products, people, and topics, plus the relations between them, each backed by the pages that mention them. You generate it from chat and export it as CSV files or a PDF report.
What's the difference between the Quick and Full graph?
Quick reads the schema.org structured data your pages already publish and builds the graph in seconds, with no AI call. Full adds an AI pass over your most important pages to catch entities the markup missed, running on your local model or your own OpenRouter key.
Do I need to re-crawl my site to get the better search?
No. The new passage index builds from crawl data you already have. There's a Rebuild index button under Settings, Site Monitoring, and the app also repairs a stale index on its own at startup.
Where does all this data live?
On your machine. The page index, the passages, and the entity store are rows in a local database, and the embedding model that powers search runs on your device. Nothing about your site has to be uploaded to build the graph.
The one who's under the desk for all of it.
