Find who sells it near you or online, ranked by how they treat people and the planet, never Amazon.

Type a product and a US zip code. You get shops near you and online retailers that sell it or something close, each ranked with its reasons shown. There are no accounts, cookies, or analytics.

Why Amazon never appears

Amazon, and every business it owns or has owned, never appears in results at any rank. That includes Whole Foods Market, Zappos, Ring, and Audible. Code removes them twice: once right after the web search, and again as the last step before results reach your browser. The language model is never the only check.

Every entry on the exclusion list cites a public source: data/blocklist.md.

How ranking works

Each result gets four scores between 0 and 1, and its rank comes from their weighted sum. The weights are listed above every set of results, and "Why this rank" on each result says in plain words what went into its score, linking each certification or concern to its source.

Relevance: 1 when the shop's listing names the product or a close equivalent, 0.5 when it matches only the product's category, otherwise 0.2. For a shop near you, the language model's judgment from the shop's name and type also counts: 1 if it judged the shop likely sells the product, 0.5 if maybe.
Ethics: starts at 0.5. Each ethics certification (B Corp, Fair Trade, worker co-op) adds 0.25, up to 1. An accepted labor or governance finding subtracts up to 0.25: less when the penalty was small, the amount isn't stated, the case is still open, or it was against a related company; small findings together subtract at most 0.25; down to 0. The ranking page has the table.
Environment: starts at 0.5. Each environmental certification (1% for the Planet, Climate Neutral) adds 0.25, up to 1. An accepted environmental finding subtracts up to 0.25 the same way; down to 0.
Proximity: for shops near you, 1 at the center of your zip area, falling to 0 at 40 km (about 25 miles). Online retailers get 0.5.

"Baseline 0.5" in a score's sources means that score started at 0.5 and nothing changed it.

Certifications come from a short curated list in which every entry cites its source. A negative finding lowers a score only when its source is on the list of accepted sources: regulators, court records, established news organizations, and recognized labor and environmental watchdogs. The language model suggests search terms, judges whether each result is a shop that sells the product and leaves out those it judges are not, and points to findings in the search results; the text shown for a finding is always the source's own title or snippet, never words the model wrote. The chain badge comes from a curated list of chains with 10 or more US stores, each entry linking its own source, Like a certification it can be disputed; unlike one, it never changes the score. Details: docs/ranking.md.

Disputing a ranking

Anyone can dispute a certification, a chain badge, a negative finding, or any part of a score, including the business itself. Every negative finding shown on a result has a "Dispute this" link that leads here or to the dispute form. How disputes are reviewed and resolved: TBD.

Privacy

Your zip code never leaves your browser. The site looks it up in a file it downloads and sends only the product, the city and state, the center of your zip area rounded to two decimal places, and its rural-urban category (used only to decide how far counts as nearby). The search service uses that center to find nearby shops, so it learns the zip area of each search; the language model's first call sees only the product, and its second sees the search results, which can name nearby towns. The site never asks for a street address. The map of nearby shops loads its images from OpenStreetMap, so OpenStreetMap sees your IP address and roughly where the search was.

What is fresh on every search, and what is kept:

Your search as a whole (the product together with where you searched from): used for that search only and never stored by this site.
Shops found, whether each likely sells the product, and their scores: looked up fresh for every search and not kept by this site.
The model's reading of a product (its category and search wording): kept in Cloudflare's storage for 30 days from the first search for it. That first reading is shared by everyone who types the same product, in any capitalization, so repeat searches for it use the same search wording. In effect it is a list of products people searched for, with no location or IP address.
Your IP address (for IPv6, its first half): kept by the rate limiter for one minute, counting searches from that address. A copy in the search service's memory stops counting after the minute and is erased when old entries are cleared or the service restarts.
The search service's log line (whether the search worked and how long it took, never the product or location): not kept, because its log storage is turned off.
Request logs at Cloudflare and GitHub Pages, which host the site: kept under each platform's own policy.
What Brave, OpenRouter, the model provider and OpenStreetMap receive for a search: kept by each under its own policy.
The last product you typed: kept in this browser tab until you close it, for you only. Never the zip.

Details: docs/privacy.md.

Coverage and quality

Results come from Brave Search's index. Shops that block search crawlers, or that the index has not reached, may be missing. Quality is measured on a fixed set of test searches: how many results really sell the product (precision), and how many good shops found by an independent search engine this site also found (recall).

Not yet measured.

Raw evidence: docs/evidence/quality. Method and limits: docs/quality.md.

Source and cost

The site is open source under the GNU Affero General Public License v3.0. What each search costs to run is recorded in docs/costs.md, and each search shows its own cost at the bottom of the page.

Zip code coordinates: US Census Bureau ZCTA Gazetteer (public domain). City and state names: GeoNames, licensed under CC BY 4.0.