Search

Normalizes with kz-text (transliteration, мкр / микр. / шағын аудан, дом / үй, подъезд / кіреберіс), then recalls from both the osm and ugc documents (never merged into one record). Metered per autocomplete session: requests sharing session_token count once; a session ends when a suggestion is resolved with getPlace / searchAddress or after 3 minutes idle.

Either q (free text), the structured fields, or category with near / bbox. House, microdistrict-house, entrance and flat numbers must match exactly (no typo tolerance). Returns building or street precision when there is no entrance evidence — never a forced entrance.

Category browse / filter (the "Cafe · Restaurant · Parking · School · Market" shortcuts):

• category restricts results to type: place documents whose category is in the list. It implies types=place: together with a types list that does not contain place the result is empty. House-number matching is skipped (addresses have no category).
• With q (or structured fields): the normal text search, filtered to those categories.
• Without q and structured fields: a pure category browse. It requires near or bbox (400 invalid_request otherwise). With near, places are sorted by distance from it (nearest first, distance_m set); with only bbox, by static rank inside the box.
• The vocabulary is the place category field as returned in SearchResult.category (today the OSM amenity / shop / … value, e.g. cafe, restaurant, parking, school, marketplace, pharmacy). Which sources feed places depends on the POI decision, so coverage varies and a valid category can return no results.

Entrance evidence is an OSM entrance node of the building or an approved crowdsourced entrance submitted for it (parent_place_id = the building's OSM id; decision D42). The result stays the OSM address (its id, source: osm, attribution); the crowdsourced entrance is listed in entrances with source: ugc and its own ugc.* id, and with the same number as an OSM node it is the one chosen (a moderated correction). A reverted entrance is no longer listed.

KNN in osm and ugc separately, assembled in the API. The nearest point is not assumed to be the user's building: building outlines and distance thresholds decide. precision says how exact the answer is.