The living index.
This is Tote’s index rendered live: the most popular verified stores as nodes, sized by catalog and lit by rank. Search for something and watch an agent do what agents actually do here — route through the index, open a store, and build a real cart.
Every point is a store that passed a live cart probe. Move your cursor to look around the index, and every search above runs the same five tools your agent gets — nothing here is mocked.
Crawl → verify → serve → rank
Crawl
A crawler walks the open web and public rankings looking for storefronts, popularity-first. Finding a store costs nothing and proves nothing.
Verify
The prober adds a real item to a real cart and checks the price charged against the price listed. Only stores that pass are ever served — and they are re-checked continuously.
Serve
Agents get the verified index as five deterministic tools. No AI runs at serve time; every call is a plain, fast, logged endpoint.
Rank
Every search, add, and checkout feeds back into store rank. Stores that convert rise; stores that break demote themselves.
Why serving is deterministic
All the intelligence is spent at index time — reading a store’s structure once, proving its cart works, learning which host actually serves it. What agents touch at serve time is boring on purpose: plain endpoints over a verified table, fast enough to sit inside an agent loop, honest enough to bet a checkout on. The animation above is not a metaphor — every hop chip is a real tool call with its real latency.