diff --git a/swh/model/to_disk.py b/swh/model/to_disk.py
new file mode 100644
--- /dev/null
+++ b/swh/model/to_disk.py
@@ -0,0 +1,109 @@
+# Copyright (C) 2018 The Software Heritage developers
+# See the AUTHORS file at the top-level directory of this distribution
+# License: GNU General Public License version 3, or any later version
+# See top-level LICENSE file for more information
+
+import itertools
+import os
+
+from swh.model.from_disk import mode_to_perms, DentryPerms
+
+SKIPPED_MESSAGE = (b'This content has not been retrieved in the '
+                   b'Software Heritage archive due to its size.')
+
+HIDDEN_MESSAGE = (b'This content is hidden.')
+
+
+def get_filtered_file_content(storage, file_data):
+    """Retrieve the file specified by file_data and apply filters for skipped
+    and missing contents.
+
+    Args:
+        storage: the storage from which to retrieve the object
+        file_data: file entry descriptor as returned by directory_ls()
+
+    Returns:
+        Bytes containing the specified content. The content will be replaced by
+        a specific message to indicate that the content could not be retrieved
+        (either due to privacy policy or because its size was too big for us to
+        archive it).
+    """
+
+    assert file_data['type'] == 'file'
+
+    if file_data['status'] == 'absent':
+        return SKIPPED_MESSAGE
+    elif file_data['status'] == 'hidden':
+        return HIDDEN_MESSAGE
+    else:
+        return list(storage.content_get([file_data['sha1']]))[0]['data']
+
+
+class DirectoryBuilder:
+    """Reconstructs the on-disk representation of a directory in the storage.
+    """
+
+    def __init__(self, storage, root, dir_id):
+        """Initialize the directory builder.
+
+        Args:
+            storage: the storage object
+            root: the path where the directory should be reconstructed
+            dir_id: the identifier of the directory in the storage
+        """
+        self.storage = storage
+        self.root = root
+        self.dir_id = dir_id
+
+    def build(self):
+        # Retrieve data from the database.
+        data = self.storage.directory_ls(self.dir_id, recursive=True)
+
+        # Split into files and directory data.
+        # TODO(seirl): also handle revision data.
+        data1, data2 = itertools.tee(data, 2)
+        dir_data = (entry['name'] for entry in data1 if entry['type'] == 'dir')
+        file_data = (entry for entry in data2 if entry['type'] == 'file')
+
+        # Recreate the directory's subtree and then the files into it.
+        self._create_tree(dir_data)
+        self._create_files(file_data)
+
+    def _create_tree(self, directory_paths):
+        """Create a directory tree from the given paths
+
+        The tree is created from `root` and each given path in
+        `directory_paths` will be created.
+
+        """
+        # Directories are sorted by depth so they are created in the
+        # right order
+        bsep = bytes(os.path.sep, 'utf8')
+        dir_names = sorted(
+            directory_paths,
+            key=lambda x: len(x.split(bsep)))
+        for dir_name in dir_names:
+            os.makedirs(os.path.join(self.root, dir_name))
+
+    def _create_files(self, file_datas):
+        """Create the files according to their status."""
+        # Then create the files
+        for file_data in file_datas:
+            path = os.path.join(self.root, file_data['name'])
+            content = get_filtered_file_content(self.storage, file_data)
+            self._create_file(path, content, file_data['perms'])
+
+    def _create_file(self, path, content, mode=0o100644):
+        """Create the given file and fill it with content."""
+        perms = mode_to_perms(mode)
+        if perms == DentryPerms.symlink:
+            os.symlink(content, path)
+        else:
+            with open(path, 'wb') as f:
+                f.write(content)
+            os.chmod(path, perms.value)
+
+    def _get_file_content(self, obj_id):
+        """Get the content of the given file."""
+        content = list(self.storage.content_get([obj_id]))[0]['data']
+        return content