HF / BUCKET ATLAS

A planet.
In a bucket.

The entire planet in one PMTiles file on Hugging Face Buckets. As you explore, your browser requests just the pieces it needs. A world of detail, a few kilobytes at a time.

Four steps from an empty bucket to a globe that streams itself.

1Bucket
Create a private bucket

Storage Buckets are S3-like object storage on the Hub: mutable, non-versioned, made for big files.
```
$ hf buckets create bucket-atlas --private
```
2Upload
Put the planet in it

One PMTiles archive holds every tile for the whole world, . It is uploaded once and never unpacked.
```
$ hf buckets cp planet-20260907.pmtiles hf://buckets/<username>/bucket-atlas/maps/
```
3Space
Read byte ranges with Bun S3

A Space holds read-only S3 credentials and forwards each Range request to the bucket with Bun S3. Nothing is downloaded or cached on disk.
```
const file = s3.file("maps/planet-20260907.pmtiles");
return fetch(file.presign(), {
  headers: { Range: req.headers.get("Range") },
});
```
4Browser
Ask only for the tiles in view

MapLibre and the PMTiles client fetch a 16 KB header, then a directory, then one small slice per tile. Pan, and it asks for more.
```
GET /maps/planet-20260907.pmtiles
Range: bytes=41287664-41329101
```

Buckets S3 guide ↗ · Bun S3 docs ↗ · Source ↗

A planet.In a bucket.

A planet.
In a bucket.