In this paper we present an approach to compute arbitrarily sized meshes of large scale environments. The meshes can be reconstructed from laser-scanned point clouds or existing high-resolution meshes. The algorithm automatically builds a level of detail hierarchy in the Cesium 3D Tiles format using an octree partition. The main contribution of this paper is a method that ensures that the generated meshes for each level-of-detail stage are computed in a consistent manner to minimize visual artifacts between different detail levels during rendering. Furthermore, both the reconstruction and simplification algorithm are designed to constrain the memory consumption, which enables to process even very large data sets on consumer-grade hardware. The export into the Cesium 3D Tiles format allows to render such large meshes efficiently in all web-based viewers that support this format. In our experiments we evaluate the method on different datasets and assess the visual quality during the rendering process and analyze the memory footprint as well as the runtime behaviour.