GeoPair · find a place from what you saw · OpenStreetMap data · GLM-5.3-Flash reads your wordshow it works

You saw some things. Where was it?

Type what you remember or what is in the photo, in any language, however you like: the shops, the stop, the tower in the distance, what was across the street, what was missing. The tool turns it into a map search and ranks the candidates.

Optional · a photo
Drop a photo here or click. What the model reads in it is added to your words.
Optional · roughly where
Try: AH + Kruidvat + tram, Amsterdam mosque, pharmacy, tram stop, NL Rewe + Apotheke + Wasserturm, Berlin Lidl + McDonald's, no railway, BE

How it works

You write what you saw. GLM-5.3-Flash rewrites it as a search: the most specific thing (a brand, a business, a landmark) becomes the anchor, the other things become constraints with a distance ("next to" is 30 m, "across the street" 60 m, "a few streets away" 400 m, "not near X" excludes), and the area becomes a city if you named one, else the country the brands and language point to. You see that plan and can change every part of it.

Then OpenStreetMap decides which places exist: every anchor in the area, kept only where the whole chain holds. Data comes from the public Overpass API, so completeness varies by country and a whole country takes 30 to 120 seconds. Finally the model ranks the candidates against everything you wrote, including what a map cannot hold (a red door, a house number, a tram line), and says what to check in Street View. When your clues are generic it says so rather than pick a winner.

Every hit links to Google Maps, Street View, Mapillary, the OpenStreetMap object and Google Earth. Exports: CSV, GeoJSON, KML. The permalink holds the whole search. Nothing is stored. Built by Henk van Ess, September 2026, as the successor of geo.graph.tips; no key needed.