GeoJSON to PMTiles

Yes — tiling, encoding and packaging all run in your browser; the file never leaves your device. The in-page inspection map reads the freshly built archive from an in-memory blob (basemap tiles are fetched over the network).

Datasets up to roughly 100 MB work well. Simplify very large datasets first, lower the max zoom, or raise the min zoom.

Yes — header bounds/center (int32 1e7 fixed-point, official reader semantics) and metadata bounds/center/vector_layers (with property column types) are written with real values, so the official MapLibre protocol adapter no longer logs "Bounds not valid" and the TileJSON bounds are no longer [0,0,0,0].

Tippecanoe is the official tiling CLI for PMTiles (v2.17+ writes PMTiles directly) and the right choice for planet-scale or scripted batch pipelines — but it is a C++ install, and on Windows it needs Cygwin or Docker. This tool does the same job in your browser: no install, no sign-up, nothing uploaded. Use it for datasets up to roughly 100 MB; reach for Tippecanoe when you need server-grade scale.

The default 0–8 suits city-to-country overviews. You can go up to z15 — enough for street-level mapping. Tiles below your min zoom are skipped to keep the archive small, so raise the min zoom or lower the max zoom when the output is too big.

Yes — the in-page inspection map reads the freshly built PMTiles from an in-memory blob through the official PMTiles protocol adapter with MapLibre (basemap tiles are fetched over the network, your data is not). Any MapLibre app can load the file the same way with the pmtiles:// protocol.

Upload the single file to R2, S3 or any static host with HTTP range-request support and render it with MapLibre GL — no tile server, no MBTiles unpacking. The header bounds and center are already in the archive, so the map frames your data on load.