What this is
Most itinerary tools either hand you a generated paragraph that reads well and falls apart on contact with an opening-hours sign, or they hand you a list and leave the arithmetic to you. This one does the arithmetic. It knows how long each place takes, when it is shut, and how long it takes to get from the third stop to the fourth — and it will tell you when the answer is "you have too many days for this city".
The guide half is ordinary and deliberately so: a page per country, region, city, district and place, ranked by importance rather than by who paid, written to be read rather than to rank. The planner half is the part that took the work.
Where the data comes from
Places, coordinates, categories and opening hours come from OpenStreetMap. Names, descriptions and translations come from Wikidata and Wikivoyage. Importance is computed from Wikipedia sitelinks, pageview counts and heritage status — three real signals, combined arithmetically, never a score invented to fill a column.
- OpenStreetMap
- Places, coordinates, categories, opening hours, administrative district boundaries. ODbL.
- Wikidata
- Names in each language, descriptions, heritage status, coordinates used to verify the OSM match.
- Wikipedia
- Pageview counts and sitelink counts, which together drive the importance ranking.
- Wikivoyage
- Source material for editorial sections, reviewed before publication.
What is deliberately absent: prices. Open data carries no reliable price level, so there is no budget filter — a "cheap" toggle with nothing behind it would mislead more than it helps. It returns when a real price source is ingested.
Why the planner runs no language model
At the moment you press the button, nothing is generated. The engine scores candidates, clusters them geographically, orders each cluster into a walking sequence and schedules it against opening hours and your day window. Same answers in, same plan out, every time.
That determinism is the point. A plan you can reproduce is a plan you can check, and two rules are enforced by tests rather than by hope: no stop is ever scheduled while its place is closed, and a day never exceeds its own window once visit time, travel time and breaks are added up.
- Language models are used at build time only, to draft editorial text that is reviewed before it is published.
- Travel times come from a routing engine, not from an estimate of "about twenty minutes".
- Where a duration is a category default rather than a measured figure, the interface says so with an approximate sign.
Coverage, honestly
Five cities are live: Istanbul, Vienna, Belgrade, Paris and Manisa. Manisa is in that list on purpose. It is small and thinly mapped — 34 places, 7% of them with recorded opening hours — which makes it the city that exercises everything the dense ones hide. If a plan for Manisa is honest, the rest follow.
Eleven content languages are published or in progress. A language appears in the switcher only when its content exists; an unpublished locale is absent rather than shown as an empty shell, and an untranslated name falls back to English with a visible marker rather than to a machine translation.
How it pays for itself
Advertising, and nothing else. There is no affiliate booking link, no paid placement in any ranking, and no sponsored place masquerading as a recommendation. Ad slots reserve their height so an unfilled ad shifts nothing, they sit between content blocks rather than inside a passage, and there is no publisher-built pop-up over what you are reading.
- A ranking is never influenced by money. The importance score has three inputs and none of them is commercial.
- The print view of an itinerary carries no advertising at all.
- Full detail is on the advertising disclosure page.