Local STL viewer and geometry inspector

Triangle count is the fastest signal for whether a slicer will accept the model. A 100 MB binary STL is roughly two million triangles.

Welded vertex count is typically about half the triangle count for a closed surface. A much larger number suggests duplicated or unwelded seams.

Bounds are reported in the source units (STL is unitless; the site treats it as millimetres by default). Check each axis against your printer build volume.

Surface area is useful for resin estimates and for spotting suspiciously flat or missing regions.

Signed volume is estimated from the triangle soup. It is meaningful for a closed, consistently oriented mesh; open meshes will under- or over-report.

Meshes above 250,000 triangles open with a vertex-clustering preview near 100,000 triangles. The preview only changes what is drawn: triangle, vertex, bounds, area and volume values still describe the full mesh. Loading the full mesh streams the geometry in independently indexed chunks from the resident Worker mesh or from the full cache, and the viewport appends each chunk without re-parsing when a resident copy is available. Chunks are ordered by distance to the current orbit target, and each chunk keeps its own bounds so off-screen chunks can be frustum-culled.

The decimated preview and, when it fits the cache budget, the full welded geometry are cached in the browser's origin-private file system (OPFS), keyed by a hash of sampled file bytes, the file size and its modification time. Full geometry up to 24 MB is stored as one pair of files; larger full meshes up to 256 MB are written as independently indexed chunks. Binary STL chunks are streamed straight from aligned file slices, so the detail cache does not need the whole welded mesh in memory. Reopening the same file restores the preview and streams cached chunks without parsing again. Beyond the chunk budget, the full mesh is re-parsed on demand. The cache stays on this device and clearing site data removes it.

Near-view detail is an optional mode for large meshes. The Worker prepares a local chunk cache, then the viewer keeps up to eight full-resolution chunks nearest the current orbit target, inside a radius that follows the camera distance over the decimated preview. Panning the target loads new nearby chunks and unloads far ones. The preview remains visible outside the detailed region; the mode requires OPFS and a 256 MB detail-cache budget per model.