diff --git a/graphique/shell.py b/graphique/shell.py
index b46a115..3fe7fe5 100644
--- a/graphique/shell.py
+++ b/graphique/shell.py
@@ -11,11 +11,17 @@
 
 import argparse
 import shutil
+from collections.abc import Iterable
 from pathlib import Path
 import pyarrow.dataset as ds
 from tqdm import tqdm  # type: ignore
 
 
+def sort_key(name: str) -> tuple:
+    """Parse sort order."""
+    return name.lstrip('-'), ('descending' if name.startswith('-') else 'ascending')
+
+
 def write_batches(scanner: ds.Scanner, base_dir: str, *partitioning: str, **options):
     """Partition dataset by batches."""
     options.update(format='parquet', partitioning=partitioning)
@@ -27,24 +33,31 @@ def write_batches(scanner: ds.Scanner, base_dir: str, *partitioning: str, **opti
             pbar.update(len(batch))
 
 
-def write_fragments(dataset: ds.Dataset, base_dir: str, **options):
+def write_fragments(dataset: ds.Dataset, base_dir: str, sorting=(), **options):
     """Rewrite partition files by fragment to consolidate."""
+    options['format'] = 'parquet'
     exprs = {Path(frag.path).parent: frag.partition_expression for frag in dataset.get_fragments()}
     offset = len(dataset.partitioning.schema)
     for path in tqdm(exprs, desc="Fragments"):
         part_dir = Path(base_dir, *path.parts[-offset:])
-        ds.write_dataset(dataset.filter(exprs[path]), part_dir, **options)
+        part = dataset.filter(exprs[path])
+        ds.write_dataset(part.sort_by(sorting) if sorting else part, part_dir, **options)
 
 
 def partition(
-    scanner: ds.Scanner, base_dir: str, *partitioning: str, fragments: bool = False, **options
+    scanner: ds.Scanner,
+    base_dir: str,
+    *partitioning: str,
+    fragments: bool = False,
+    sort: Iterable[str] = (),
+    **options,
 ):
     """Partition dataset by keys."""
     temp = Path(base_dir) / 'temp'
     write_batches(scanner, str(temp), *partitioning)
     dataset = ds.dataset(temp, partitioning='hive')
-    if fragments:
-        write_fragments(dataset, base_dir, **options)
+    if fragments or sort:
+        write_fragments(dataset, base_dir, tuple(map(sort_key, sort)), **options)
     else:
         options.update(partitioning_flavor='hive', existing_data_behavior='overwrite_or_ignore')
         with tqdm(desc="Partitions"):
@@ -57,8 +70,9 @@ def partition(
 parser.add_argument('dest', help="destination path")
 parser.add_argument('partitioning', nargs='+', help="partition keys")
 parser.add_argument('--fragments', action='store_true', help="iterate over fragments")
+parser.add_argument('--sort', nargs='*', default=(), help="sort keys; will load fragments")
 
 if __name__ == '__main__':
     args = parser.parse_args()
     dataset = ds.dataset(args.src, partitioning='hive')
-    partition(dataset, args.dest, *args.partitioning, fragments=args.fragments)
+    partition(dataset, args.dest, *args.partitioning, fragments=args.fragments, sort=args.sort)