There are no tiles. All geometry is loaded and rendered directly from an SQLite database on every render.

Geometry is stored as TWKB (Tiny Well-Known Binary) to reduce storage and transport size. During decoding, they do clever work using aggregate functions and reusing buffers across rows to reduce allocations.

There is real potential in the tech, but unfortunately little momentum behind it.

▲

palata an hour ago | parent [-]

> There are no tiles. All geometry is loaded and rendered directly from an SQLite database on every render.

Which can be done with tiles. Or maybe I don't understand what you mean by "tiles"? What do you describe as "tiles"?

	▲	mamaar 34 minutes ago \| parent [-]
		I would define a tile as a slice of geometries, not whole features. The difference with Tileless' approach, is that they load whole features from the database and don't split them into tiles. So if a feature extends outside the current view, they would load the whole geometry rather than the intersection of the tile's extent and the geometry. Vector tiles are optimized for concurrent downloads and browser / CDN caching and doing a good job of that.