diff --git a/docs/datastore-narrative_example.rst b/docs/datastore-narrative_example.rst
new file mode 100644
index 000000000000..28110831a8d6
--- /dev/null
+++ b/docs/datastore-narrative_example.rst
@@ -0,0 +1,382 @@
+Cloud Datastore Narrative / Example Application
+===============================================
+
+Overview
+--------
+
+In order to give a better feel for how a Python application might use the
+:mod:`gcloud.datastore` API, let's look at building an example application
+which stores its data using the API.  In order to focus on the API, the
+sample application will be built as a set of command-line scripts.
+
+For our example, let's build an employee expense reporting system.  An example
+set of interactions might look like the following scenario.
+
+Returning from a trip to the Bay Area, Sally creates a CSV file,
+``expenses-20140901.csv``, containing one row for each line item in her
+expense report:
+
+.. code-block:: none
+
+   "Date","Vendor","Type","Quantity","Price","Memo"
+   "2014-08-26","United Airlines","Travel",1,425.00,"Airfaire, IAD <-> SFO"
+   "2014-08-27","Yellow Cab","Travel",32.00,"Taxi to IAD"
+   ...
+
+Sally then submits her expense report from the command line using our
+:program:`submit_expenses` script (see :ref:`create-expense-report`):
+
+.. code-block:: bash
+
+   $ submit_expenses create --employee-id=sally --description="Frotz project kickoff, San Jose" expenses-20140901.csv
+   Processed 15 rows.
+   Created report: sally/expenses-20140901
+
+Sally can list all her submitted expense reports using our
+:program:`review_expenses` script (see :ref:`list-expense-reports`):
+
+.. code-block:: bash
+
+   $ review_expenses list --employee-id=sally
+
+   "Employee ID", "Report ID","Created","Updated","Description","Status","Memo"
+   "sally","expenses-2013-11-19","2013-12-01","2013-12-01","Onsite Training, Mountain View","Paid","Check #3715"
+   "sally","expenses-2014-04-19","2014-04-21","2014-04-22","PyCon 2014, Montreal","Paid","Check #3992"
+   "sally","expenses-2014-09-01","2014-09-04","2014-09-04","Frotz project kickoff, San Jose","pending",""
+
+Sally can review a submitted expense report using the its ID
+(see :ref:`show-expense-report`):
+
+.. code-block:: bash
+
+   $ review_expenses show sally expenses-20140901
+   Employee-ID: sally
+   Report-ID: expenses-20140901
+   Report-Status: pending
+   Created: 2014-09-04
+   Updated: 2014-09-04
+   Description: Frotz project kickoff, San Jose
+
+   "Date","Vendor","Type","Quantity","Price","Memo"
+   "2014-08-26","United Airlines","Travel",1,425.00,"Airfaire, IAD <-> SFO"
+   "2014-08-27","Yellow Cab","Travel",32.00,"Taxi to IAD"
+   ...
+
+While in "pending" status, Sally can edit the CSV and resubmit it
+(see :ref:`update-expense-report`):
+
+.. code-block:: bash
+
+   $ submit_expenses update expenses-20140901.csv
+   Updated report: sally/expenses-20140901
+   Processed 15 rows.
+
+While it remains in "pending" status, Sally can also delete the report
+(see :ref:`delete-expense-report`):
+
+.. code-block:: bash
+
+   $ submit_expenses delete expenses-20140901
+   Deleted report: sally/expenses-20140901
+   Removed 15 items.
+
+Sally's boss, Pat, can review all open expense reports
+(see :ref:`list-expense-reports`):
+
+.. code-block:: bash
+
+   $ review_expenses list --status=pending
+
+   "Employee ID","Report ID","Created","Updated","Description","Status","Memo"
+   "sally","expenses-2014-09-01","2014-09-04","2014-09-04","Frotz project kickoff, San Jose","pending",""
+
+
+Pat can download Sally's report
+(see :ref:`show-expense-report`):
+
+.. code-block:: bash
+
+   $ review_expenses show sally expenses-20140901
+   Report-ID: sally/expenses-20140901
+   Report-Status: pending
+   Employee-ID: sally
+   Description: Frotz project kickoff, San Jose
+
+   "Date","Vendor","Type","Quantity","Price","Memo"
+   "2014-08-26","United Airlines","Travel",1,425.00,"Airfaire, IAD <-> SFO"
+   "2014-08-27","Yellow Cab","Travel",32.00,"Taxi to IAD"
+
+Pat can approve Sally's expense report
+(see :ref:`approve-expense-report`):
+
+.. code-block:: bash
+
+   $ review_expenses approve --check-number=4093 sally expenses-20140901
+   Approved, report: sally/expenses-20140901, check #4093
+
+or reject it
+(see :ref:`reject-expense-report`):
+
+.. code-block:: bash
+
+   $ review_expenses reject --reason="Travel not authorized by client" sally expenses-20140901
+   Rejected, report: sally/expenses-20140901, reason: Travel not authorized by client
+
+Connecting to the API Dataset
+-----------------------------
+
+The sample application uses a utility function, :func:`expenses._get_dataset`,
+to set up the connection.
+
+.. literalinclude:: examples/expenses/expenses/__init__.py
+   :pyobject: _get_dataset
+
+Thie function expects three environment variables to be set up, using
+your project's
+`OAuth2 API credentials <https://developers.google.com/console/help/new/#generatingoauth2>`_:
+
+- :envvar:`GCLOUD_TESTS_DATASET_ID` is your Google API Project ID
+- :envvar:`GCLOUD_TESTS_CLIENT_EMAIL` is your Google API email-address
+- :envvar:`GCLOUD_TESTS_TESTS_KEY_FILE` is the filesystem path to your
+  Google API private key.
+
+.. _create-expense-report:
+
+Creating a New Expense Report
+-----------------------------
+
+In the sample application, the ``create`` subcommand of the
+:program:`submit_expenses` script drives a function,
+:func:`expenses.create_report`:
+
+.. literalinclude:: examples/expenses/expenses/__init__.py
+   :pyobject: create_report
+   :linenos:
+
+After connecting to the dataset via :func:`expenses._get_dataset` (line 2),
+:func:`expenses.create_report` starts a transaction (line 3) to ensure that
+all changes are performed atomically.  It then checks that no report exists
+already for the given employee ID and report ID, raising an exception if so
+(lines 4-5).  It then  delegates most of the work to the
+:func:`expenses._upsert_report` utility function (line 6), finally setting
+metadata on the report itself (lines 7-11).
+
+
+.. literalinclude:: examples/expenses/expenses/__init__.py
+   :pyobject: _upsert_report
+   :linenos:
+
+The :func:`expenses._upsert_report` function: in turn delegates to
+:func:`expenses._get_employee`, :func:`expenses._get_report`, and
+:func:`expenses._purge_report_items` to ensure that the employee and report
+exist, and that the report contains no items (lines 2-4).  It then
+iterates over the rows from the CSV file, creating an item for each row
+(lines 5-13), finally returning the populated report object (line 14).
+
+.. literalinclude:: examples/expenses/expenses/__init__.py
+   :pyobject: _get_employee
+   :linenos:
+
+The :func:`expenses._get_employee` function: looks up an employee (lines 2-3).
+
+.. note:: Employee entities have no "parent" object: they exist at the "top"
+          level.
+
+If the employee entity does not exist, and the caller requests it, the
+function creates a new employee entity and saves it.
+
+.. literalinclude:: examples/expenses/expenses/__init__.py
+   :pyobject: _get_report
+   :linenos:
+
+The :func:`expenses._get_employee` function: looks up an expense report
+using an `"ancestor" query
+<https://cloud.google.com/datastore/docs/concepts/queries#Datastore_Ancestor_queries>`_
+(lines 2-3).
+
+.. note:: Each expense report entity is expected to have an employee entity
+          as its "parent".
+
+If the expense report entity does not exist, and the caller requests it, the
+function creates a new expense report entity and saves it.
+
+.. literalinclude:: examples/expenses/expenses/__init__.py
+   :pyobject: _purge_report_items
+   :linenos:
+
+The :func:`expenses._purge_report_items` function: delegates to
+:func:`expenses._fetch_report_items` to find expense item entities contained
+within the given report (line 4), and deletes them (line 5).  It returns
+a count of the deleted items.
+
+.. literalinclude:: examples/expenses/expenses/__init__.py
+   :pyobject: _fetch_report_items
+   :linenos:
+
+The :func:`expenses._purge_report_items` function: performs an "ancestor"
+query (lines 2-3) to find expense item entities contained within a given
+expense report.
+
+.. _update-expense-report:
+
+Updating an Existing Expense Report
+-----------------------------------
+
+In the sample application, the ``update`` subcommand of the
+:program:`submit_expenses` script drives a function,
+:func:`expenses.update_report`:
+
+.. literalinclude:: examples/expenses/expenses/__init__.py
+   :pyobject: update_report
+   :linenos:
+
+After connecting to the dataset via :func:`expenses._get_dataset` (line 2),
+:func:`expenses.update_report` starts a transaction (line 3) to ensure that
+all changes are performed atomically.  It then checks that a report *does*
+exist already for the given employee ID and report ID, and that it is in
+``pending`` status, raising an exception if not (lines 4-5).  It then
+delegates most of the work to the :func:`expenses._upsert_report` utility
+function (line 6), finally updating metadata on the report itself (lines 7-11).
+
+.. _delete-expense-report:
+
+Deleting an Existing Expense Report
+-----------------------------------
+
+In the sample application, the ``delete`` subcommand of the
+:program:`submit_expenses` script drives a function,
+:func:`expenses.delete_report`:
+
+.. literalinclude:: examples/expenses/expenses/__init__.py
+   :pyobject: delete_report
+   :linenos:
+
+After connecting to the dataset via :func:`expenses._get_dataset` (line 2),
+:func:`expenses.delete_report` starts a transaction (line 3) to ensure that
+all changes are performed atomically.  It then checks that a report *does*
+exist already for the given employee ID and report ID (lines 4-6), and that
+it is in ``pending`` status (lines 7-8), raising an exception if either is
+false.
+
+.. note::
+
+   The function takes a ``force`` argument:  if true, it will delete the
+   report even if it is not in ``pending`` status.
+
+The function then delegates to :func:`expenses._purge_report_items` to
+delete expense item entities contained in the report (line 9), and then
+deletes the report itself (line 10).  Finally, it returns a count of the
+deleted items.
+
+.. _list-expense-reports:
+
+Listing Expense Reports
+-----------------------
+
+In the sample application, the ``list`` subcommand of the
+:program:`review_expenses` script drives a function,
+:func:`expenses.list_reports`:
+
+.. literalinclude:: examples/expenses/expenses/__init__.py
+   :pyobject: list_reports
+   :linenos:
+
+After connecting to the dataset via :func:`expenses._get_dataset` (line 2),
+:func:`expenses.list_reports` creates a :class:`~gcloud.dataset.query.Query`
+instance, limited to entities of kind, ``Expense Report`` (line 3), and
+applies filtering based on the passed criteria:
+
+- If ``employee_id`` is passed, it adds an "ancestor" filter to
+  restrict the results to expense reports contained in the given employee
+  (lines 4-6).
+
+- If ``status`` is passed, it adds an "attribute" filter to
+  restrict the results to expense reports which have that status (lines 7-8).
+
+.. note::
+
+   The function does *not* set up a transaction, as it uses only
+   "read" operations on the API.
+
+Finally, the function fetches the expense report entities returned by
+the query and iterates over them, passing each to
+:func:`expenses._report_info` and yielding the mapping it returns.
+report (lines 9-10).
+
+.. literalinclude:: examples/expenses/expenses/__init__.py
+   :pyobject: _report_info
+   :linenos:
+
+The :func:`expenses._report_info` utility function uses the expense report
+entity's key to determine the report's employee ID (line 3), and its
+report ID (line 4).  It then uses these values and the entityy's properties
+to generate and return a mapping describing the report (lines 5-22).
+
+.. _show-expense-report:
+
+Showing an Expense Report
+-------------------------
+
+In the sample application, the ``show`` subcommand of the
+:program:`review_expenses` script drives a function,
+:func:`expenses.get_report_info`:
+
+.. literalinclude:: examples/expenses/expenses/__init__.py
+   :pyobject: get_report_info
+   :linenos:
+
+After connecting to the dataset via :func:`expenses._get_dataset` (line 2),
+:func:`expenses.get_report_info` uses :func:`exenses._get_report` to fetch
+the expense report entity for the given employee ID and report ID (line 3),
+raising an exeception if the report does not exist (line 4):
+
+.. note::
+
+   The function does *not* set up a transaction, as it uses only
+   "read" operations on the API.
+
+The function delegates to :func:`expenses._report_info` to get a mapping
+describing the report (line 6), and then delegates to
+:func:`expenses._fetch_report_items` to retrieve information about the
+expense item entities contained in the report (line 7).  Finally, the
+function returns the mapping.
+
+.. _approve-expense-report:
+
+Approving an Expense Report
+---------------------------
+
+In the sample application, the ``approve`` subcommand of the
+:program:`review_expenses` script drives a function,
+:func:`expenses.approve_report`:
+
+.. literalinclude:: examples/expenses/expenses/__init__.py
+   :pyobject: approve_report
+   :linenos:
+
+After connecting to the dataset via :func:`expenses._get_dataset` (line 2),
+:func:`expenses.approve_report` starts a transaction (line 3) to ensure that
+all changes are performed atomically.  It then checks that a report *does*
+exist already for the given employee ID and report ID, and that it is in
+``pending`` status, raising an exception if not (lines 4-5).  It then updates
+the status and other metadata on the report itself (lines 9-12).
+
+.. _reject-expense-report:
+
+Rejecting an Expense Report
+---------------------------
+
+In the sample application, the ``reject`` subcommand of the
+:program:`review_expenses` script drives a function,
+:func:`expenses.reject_report`:
+
+.. literalinclude:: examples/expenses/expenses/__init__.py
+   :pyobject: reject_report
+   :linenos:
+
+After connecting to the dataset via :func:`expenses._get_dataset` (line 2),
+:func:`expenses.approve_report` starts a transaction (line 3) to ensure that
+all changes are performed atomically.  It then checks that a report *does*
+exist already for the given employee ID and report ID, and that it is in
+``pending`` status, raising an exception if not (lines 4-5).  It then updates
+the status and other metadata on the report itself (lines 9-12).
diff --git a/docs/examples/expenses/expenses-20140901.csv b/docs/examples/expenses/expenses-20140901.csv
new file mode 100644
index 000000000000..47ae47fad965
--- /dev/null
+++ b/docs/examples/expenses/expenses-20140901.csv
@@ -0,0 +1,3 @@
+"Date","Vendor","Type","Quantity","Price","Memo"
+"2014-08-26","United Airlines","Travel",1,425.00,"Airfaire, IAD <-> SFO"
+"2014-08-27","Yellow Cab","Travel",1,32.00,"Taxi to IAD"
diff --git a/docs/examples/expenses/expenses/__init__.py b/docs/examples/expenses/expenses/__init__.py
new file mode 100644
index 000000000000..0944c369fcb7
--- /dev/null
+++ b/docs/examples/expenses/expenses/__init__.py
@@ -0,0 +1,195 @@
+import datetime
+import os
+
+from gcloud import datastore
+from gcloud.datastore.key import Key
+from gcloud.datastore.entity import Entity
+from gcloud.datastore.query import Query
+
+
+class DuplicateReport(Exception):
+    """Attempt to create a report which already exists."""
+
+
+class NoSuchReport(Exception):
+    """Attempt to update / delete a report which does not already exist."""
+
+
+class BadReportStatus(Exception):
+    """Attempt to update / delete an already-approved/rejected report."""
+
+
+def _get_dataset():
+    client_email = os.environ['GCLOUD_TESTS_CLIENT_EMAIL']
+    private_key_path = os.environ['GCLOUD_TESTS_KEY_FILE']
+    dataset_id = os.environ['GCLOUD_TESTS_DATASET_ID']
+    return datastore.get_dataset(dataset_id, client_email, private_key_path)
+
+
+def _get_employee(dataset, employee_id, create=True):
+    key = Key(path=[
+        {'kind': 'Employee', 'name': employee_id},
+        ])
+    employee = dataset.get_entity(key)
+    if employee is None and create:
+        employee = dataset.entity('Employee').key(key)
+        employee.save()
+    return employee
+
+
+def _get_report(dataset, employee_id, report_id, create=True):
+    key = Key(path=[
+        {'kind': 'Employee', 'name': employee_id},
+        {'kind': 'Expense Report', 'name': report_id},
+        ])
+    report = dataset.get_entity(key)
+    if report is None and create:
+        report = dataset.entity('Report').key(key)
+        report.save()
+    return report
+
+
+def _fetch_report_items(dataset, report):
+    query = Query('Expense Item', dataset)
+    for item in query.ancestor(report.key()).fetch():
+        yield item
+
+
+def _report_info(report):
+    path = report.key().path()
+    employee_id = path[0]['name']
+    report_id = path[1]['name']
+    created = report['created'].strftime('%Y-%m-%d')
+    updated = report['updated'].strftime('%Y-%m-%d')
+    status = report['status']
+    if status == 'paid':
+        memo = report['check_number']
+    elif status == 'rejected':
+        memo = report['rejected_reason']
+    else:
+        memo = ''
+    return {
+        'employee_id': employee_id,
+        'report_id': report_id,
+        'created': created,
+        'updated': updated,
+        'status': status,
+        'description': report.get('description', ''),
+        'memo': memo,
+        }
+
+
+def _purge_report_items(dataset, report):
+    # Delete any existing items belonging to report
+    count = 0
+    for item in _fetch_report_items(dataset, report):
+        item.delete()
+        count += 1
+    return count
+
+
+def _upsert_report(dataset, employee_id, report_id, rows):
+    _get_employee(dataset, employee_id)  # force existence
+    report = _get_report(dataset, employee_id, report_id)
+    _purge_report_items(dataset, report)
+    # Add items based on rows.
+    report_path = report.key().path()
+    for i, row in enumerate(rows):
+        path = report_path + [{'kind': 'Expense Item', 'id': i + 1}]
+        key = Key(path=path)
+        item = Entity(dataset, 'Expense Item').key(key)
+        for k, v in row.items():
+            item[k] = v
+        item.save()
+    return report
+
+
+def list_reports(employee_id=None, status=None):
+    dataset = _get_dataset()
+    query = Query('Expense Report', dataset)
+    if employee_id is not None:
+        key = Key(path=[{'kind': 'Employee', 'name': employee_id}])
+        query = query.ancestor(key)
+    if status is not None:
+        query = query.filter('status =', status)
+    for report in query.fetch():
+        yield _report_info(report)
+
+
+def get_report_info(employee_id, report_id):
+    dataset = _get_dataset()
+    report = _get_report(dataset, employee_id, report_id, False)
+    if report is None:
+        raise NoSuchReport()
+    info = _report_info(report)
+    info['items'] = [dict(x) for x in _fetch_report_items(dataset, report)]
+    return info
+
+
+def create_report(employee_id, report_id, rows, description):
+    dataset = _get_dataset()
+    with dataset.transaction():
+        if _get_report(dataset, employee_id, report_id, False) is not None:
+            raise DuplicateReport()
+        report = _upsert_report(dataset, employee_id, report_id, rows)
+        report['status'] = 'pending'
+        if description is not None:
+            report['description'] = description
+        report['created'] = report['updated'] = datetime.datetime.utcnow()
+        report.save()
+
+
+def update_report(employee_id, report_id, rows, description):
+    dataset = _get_dataset()
+    with dataset.transaction():
+        report = _get_report(dataset, employee_id, report_id, False)
+        if report is None:
+            raise NoSuchReport()
+        if report['status'] != 'pending':
+            raise BadReportStatus(report['status'])
+        _upsert_report(dataset, employee_id, report_id, rows)
+        if description is not None:
+            report['description'] = description
+        report['updated'] = datetime.datetime.utcnow()
+        report.save()
+
+
+def delete_report(employee_id, report_id, force):
+    dataset = _get_dataset()
+    with dataset.transaction():
+        report = _get_report(dataset, employee_id, report_id, False)
+        if report is None:
+            raise NoSuchReport()
+        if report['status'] != 'pending' and not force:
+            raise BadReportStatus(report['status'])
+        count = _purge_report_items(dataset, report)
+        report.delete()
+    return count
+
+
+def approve_report(employee_id, report_id, check_number):
+    dataset = _get_dataset()
+    with dataset.transaction():
+        report = _get_report(dataset, employee_id, report_id, False)
+        if report is None:
+            raise NoSuchReport()
+        if report['status'] != 'pending':
+            raise BadReportStatus(report['status'])
+        report['updated'] = datetime.datetime.utcnow()
+        report['status'] = 'paid'
+        report['check_number'] = check_number
+        report.save()
+
+
+def reject_report(employee_id, report_id, reason):
+    dataset = _get_dataset()
+    with dataset.transaction():
+        report = _get_report(dataset, employee_id, report_id, False)
+        if report is None:
+            raise NoSuchReport()
+        if report['status'] != 'pending':
+            raise BadReportStatus(report['status'])
+        report['updated'] = datetime.datetime.utcnow()
+        report['status'] = 'rejected'
+        report['reason'] = reason
+        report.save()
diff --git a/docs/examples/expenses/expenses/scripts/__init__.py b/docs/examples/expenses/expenses/scripts/__init__.py
new file mode 100644
index 000000000000..5bb534f795ae
--- /dev/null
+++ b/docs/examples/expenses/expenses/scripts/__init__.py
@@ -0,0 +1 @@
+# package
diff --git a/docs/examples/expenses/expenses/scripts/review_expenses.py b/docs/examples/expenses/expenses/scripts/review_expenses.py
new file mode 100644
index 000000000000..cf6366d452d2
--- /dev/null
+++ b/docs/examples/expenses/expenses/scripts/review_expenses.py
@@ -0,0 +1,296 @@
+import csv
+import optparse
+import os
+import textwrap
+import sys
+
+from .. import NoSuchReport
+from .. import approve_report
+from .. import get_report_info
+from .. import list_reports
+from .. import reject_report
+
+
+class InvalidCommandLine(ValueError):
+    pass
+
+
+class NotACommand(object):
+    def __init__(self, bogus):
+        self.bogus = bogus
+
+    def __call__(self):
+        raise InvalidCommandLine('Not a command: %s' % self.bogus)
+
+
+def _get_csv(args):
+    try:
+        csv_file, = args
+    except:
+        raise InvalidCommandLine('Specify one CSV file')
+    csv_file = os.path.abspath(os.path.normpath(csv_file))
+    if not os.path.exists(csv_file):
+        raise InvalidCommandLine('Invalid CSV file: %s' % csv_file)
+    with open(csv_file) as f:
+        return csv_file, list(csv.DictReader(f))
+
+
+class ListReports(object):
+    """List expense reports according to specified criteria.
+    """
+    def __init__(self, submitter, *args):
+        self.submitter = submitter
+        args = list(args)
+        parser = optparse.OptionParser(
+            usage="%prog [OPTIONS]")
+
+        parser.add_option(
+            '-e', '--employee-id',
+            action='store',
+            dest='employee_id',
+            default=None,
+            help="ID of the employee whose expense reports to list")
+
+        parser.add_option(
+            '-s', '--status',
+            action='store',
+            dest='status',
+            default=None,
+            help="Status of expense reports to list")
+
+        options, args = parser.parse_args(args)
+        self.employee_id = options.employee_id
+        self.status = options.status
+
+    def __call__(self):
+        _cols = [
+            ('employee_id', 'Employee ID'),
+            ('report_id', 'Report ID'),
+            ('created', 'Created'),
+            ('updated', 'Updated'),
+            ('description', 'Description'),
+            ('status', 'Status'),
+            ('memo', 'Memo'),
+            ]
+        writer = csv.writer(sys.stdout)
+        writer.writerow([x[1] for x in _cols])
+        for report in list_reports(self.employee_id, self.status):
+            writer.writerow([report[x[0]] for x in _cols])
+
+
+class ShowReport(object):
+    """Dump the contents of a given expense report.
+    """
+    def __init__(self, submitter, *args):
+        self.submitter = submitter
+        args = list(args)
+        parser = optparse.OptionParser(
+            usage="%prog [OPTIONS] EMPLOYEE_ID REPORT_ID")
+
+        _, args = parser.parse_args(args)
+        try:
+            self.employee_id, self.report_id, = args
+        except:
+            raise InvalidCommandLine('Specify employee ID, report ID')
+
+    def __call__(self):
+        _cols = ['Date', 'Vendor', 'Type', 'Quantity', 'Price', 'Memo']
+        try:
+            info = get_report_info(self.employee_id, self.report_id)
+        except NoSuchReport:
+            self.submitter.blather("No such report: %s/%s"
+                                   % (self.employee_id, self.report_id))
+        else:
+            self.submitter.blather("Employee-ID: %s" % info['employee_id'])
+            self.submitter.blather("Report-ID: %s" % info['report_id'])
+            self.submitter.blather("Report-Status: %s" % info['status'])
+            self.submitter.blather("Created: %s" % info['created'])
+            self.submitter.blather("Updated: %s" % info['updated'])
+            self.submitter.blather("Description: %s" % info['description'])
+            self.submitter.blather("")
+            writer = csv.writer(sys.stdout)
+            writer.writerow([x for x in _cols])
+            for item in info['items']:
+                writer.writerow([item[x] for x in _cols])
+
+
+class ApproveReport(object):
+    """Approve a given expense report.
+    """
+    def __init__(self, submitter, *args):
+        self.submitter = submitter
+        args = list(args)
+        parser = optparse.OptionParser(
+            usage="%prog [OPTIONS] EMPLOYEE_ID REPORT_ID")
+
+        parser.add_option(
+            '-c', '--check-number',
+            action='store',
+            dest='check_number',
+            default='',
+            help="Check number issued to pay the expense report")
+
+        options, args = parser.parse_args(args)
+        try:
+            self.employee_id, self.report_id, = args
+        except:
+            raise InvalidCommandLine('Specify employee ID, report ID')
+        self.check_number = options.check_number
+
+    def __call__(self):
+        approve_report(self.employee_id, self.report_id, self.check_number)
+        memo = ('' if self.check_number is None else
+                ', check #%s' % self.check_number)
+        self.submitter.blather("Approved report: %s/%s%s" %
+                               (self.employee_id, self.report_id, memo))
+
+
+class RejectReport(object):
+    """Reject a given expense report.
+    """
+    def __init__(self, submitter, *args):
+        self.submitter = submitter
+        args = list(args)
+        parser = optparse.OptionParser(
+            usage="%prog [OPTIONS] EMPLOYEE_ID REPORT_ID")
+
+        parser.add_option(
+            '-r', '--reason',
+            action='store',
+            dest='reason',
+            default=None,
+            help="Reason for rejecting the expense report")
+
+        options, args = parser.parse_args(args)
+        try:
+            self.employee_id, self.report_id, = args
+        except:
+            raise InvalidCommandLine('Specify employee ID, report ID')
+        self.reason = options.reason
+
+    def __call__(self):
+        reject_report(self.employee_id, self.report_id, self.reason)
+        memo = ('' if self.reason is None else ', reason: %s' % self.reason)
+        self.submitter.blather("Rejected report: %s/%s%s" %
+                               (self.employee_id, self.report_id, memo))
+
+
+_COMMANDS = {
+    'list': ListReports,
+    'show': ShowReport,
+    'approve': ApproveReport,
+    'reject': RejectReport,
+}
+
+
+def get_description(command):
+    klass = _COMMANDS[command]
+    doc = getattr(klass, '__doc__', '')
+    if doc is None:
+        return ''
+    return ' '.join([x.lstrip() for x in doc.split('\n')])
+
+
+class ReviewExpenses(object):
+    """ Driver for the :command:`review_expenses` command-line script.
+    """
+    def __init__(self, argv=None, logger=None):
+        self.commands = []
+        if logger is None:
+            logger = self._print
+        self.logger = logger
+        self.parse_arguments(argv)
+
+    def parse_arguments(self, argv=None):
+        """ Parse subcommands and their options from an argv list.
+        """
+        # Global options (not bound to sub-command)
+        mine = []
+        queue = [(None, mine)]
+
+        def _recordCommand(arg):
+            if arg is not None:
+                queue.append((arg, []))
+
+        for arg in argv:
+            if arg in _COMMANDS:
+                _recordCommand(arg)
+            else:
+                queue[-1][1].append(arg)
+
+        _recordCommand(None)
+
+        usage = ("%prog [GLOBAL_OPTIONS] "
+                 "[command [COMMAND_OPTIONS]* [COMMAND_ARGS]]")
+        parser = optparse.OptionParser(usage=usage)
+
+        parser.add_option(
+            '-s', '--help-commands',
+            action='store_true',
+            dest='help_commands',
+            help="Show command help")
+
+        parser.add_option(
+            '-q', '--quiet',
+            action='store_const', const=0,
+            dest='verbose',
+            help="Run quietly")
+
+        parser.add_option(
+            '-v', '--verbose',
+            action='count',
+            dest='verbose',
+            default=1,
+            help="Increase verbosity")
+
+        options, args = parser.parse_args(mine)
+
+        self.options = options
+
+        for arg in args:
+            self.commands.append(NotACommand(arg))
+            options.help_commands = True
+
+        if options.help_commands:
+            keys = sorted(_COMMANDS.keys())
+            self.error('Valid commands are:')
+            for x in keys:
+                self.error(' %s' % x)
+                doc = get_description(x)
+                if doc:
+                    self.error(textwrap.fill(doc,
+                                             initial_indent='    ',
+                                             subsequent_indent='    '))
+            return
+
+        for command_name, args in queue:
+            if command_name is not None:
+                command = _COMMANDS[command_name](self, *args)
+                self.commands.append(command)
+
+    def __call__(self):
+        """ Invoke sub-commands parsed by :meth:`parse_arguments`.
+        """
+        if not self.commands:
+            raise InvalidCommandLine('No commands specified')
+
+        for command in self.commands:
+            command()
+
+    def _print(self, text):  # pragma NO COVERAGE
+        sys.stdout.write('%s/n' % text)
+
+    def error(self, text):
+        self.logger(text)
+
+    def blather(self, text, min_level=1):
+        if self.options.verbose >= min_level:
+            self.logger(text)
+
+
+def main(argv=sys.argv[1:]):
+    try:
+        ReviewExpenses(argv)()
+    except InvalidCommandLine as e:  # pragma NO COVERAGE
+        sys.stdout.write('%s\n' % (str(e)))
+        sys.exit(1)
diff --git a/docs/examples/expenses/expenses/scripts/submit_expenses.py b/docs/examples/expenses/expenses/scripts/submit_expenses.py
new file mode 100644
index 000000000000..e1d74dd9b1c5
--- /dev/null
+++ b/docs/examples/expenses/expenses/scripts/submit_expenses.py
@@ -0,0 +1,282 @@
+import csv
+import optparse
+import os
+import textwrap
+import sys
+
+
+from .. import BadReportStatus
+from .. import DuplicateReport
+from .. import NoSuchReport
+from .. import create_report
+from .. import delete_report
+from .. import update_report
+
+
+class InvalidCommandLine(ValueError):
+    pass
+
+
+class NotACommand(object):
+    def __init__(self, bogus):
+        self.bogus = bogus
+
+    def __call__(self):
+        raise InvalidCommandLine('Not a command: %s' % self.bogus)
+
+
+def _get_csv(args):
+    try:
+        csv_file, = args
+    except:
+        raise InvalidCommandLine('Specify one CSV file')
+    csv_file = os.path.abspath(os.path.normpath(csv_file))
+    if not os.path.exists(csv_file):
+        raise InvalidCommandLine('Invalid CSV file: %s' % csv_file)
+    with open(csv_file) as f:
+        return csv_file, list(csv.DictReader(f))
+
+
+class _Command(object):
+    """Base class for create / update commands.
+    """
+    def __init__(self, submitter, *args):
+        self.submitter = submitter
+        args = list(args)
+        parser = optparse.OptionParser(
+            usage="%prog [OPTIONS] CSV_FILE")
+
+        parser.add_option(
+            '-e', '--employee-id',
+            action='store',
+            dest='employee_id',
+            default=os.getlogin(),
+            help="ID of employee submitting the expense report")
+
+        parser.add_option(
+            '-r', '--report-id',
+            action='store',
+            dest='report_id',
+            default=None,
+            help="ID of the expense report to update")
+
+        parser.add_option(
+            '-d', '--description',
+            action='store',
+            dest='description',
+            default='',
+            help="Short description of the expense report")
+
+        options, args = parser.parse_args(args)
+        self.employee_id = options.employee_id
+        self.report_id = options.report_id
+        self.description = options.description
+        self.filename, self.rows = _get_csv(args)
+        if self.report_id is None:
+            fn = os.path.basename(self.filename)
+            base, _ = os.path.splitext(fn)
+            self.report_id = base
+
+
+class CreateReport(_Command):
+    """Create a new expense report from a CSV file.
+    """
+    def __call__(self):
+        try:
+            create_report(self.employee_id, self.report_id, self.rows,
+                          self.description)
+        except DuplicateReport:
+            self.submitter.blather("Report already exists: %s/%s"
+                                   % (self.employee_id, self.report_id))
+        else:
+            self.submitter.blather("Created report: %s/%s"
+                                   % (self.employee_id, self.report_id))
+            self.submitter.blather("Processed %d rows." % len(self.rows))
+
+
+class UpdateReport(_Command):
+    """Update an existing expense report from a CSV file.
+    """
+    def __call__(self):
+        try:
+            update_report(self.employee_id, self.report_id, self.rows,
+                          self.description)
+        except NoSuchReport:
+            self.submitter.blather("No such report: %s/%s"
+                                   % (self.employee_id, self.report_id))
+        except BadReportStatus as e:
+            self.submitter.blather("Invalid report status: %s/%s, %s"
+                                   % (self.employee_id, self.report_id,
+                                      str(e)))
+        else:
+            self.submitter.blather("Updated report: %s/%s"
+                                   % (self.employee_id, self.report_id))
+            self.submitter.blather("Processed %d rows." % len(self.rows))
+
+
+class DeleteReport(object):
+    """Delete an existing expense report.
+    """
+    def __init__(self, submitter, *args):
+        self.submitter = submitter
+        args = list(args)
+        parser = optparse.OptionParser(
+            usage="%prog [OPTIONS] REPORT_ID")
+
+        parser.add_option(
+            '-e', '--employee-id',
+            action='store',
+            dest='employee_id',
+            default=os.getlogin(),
+            help="ID of employee owning the expense report")
+
+        parser.add_option(
+            '-f', '--force',
+            action='store_true',
+            dest='force',
+            default=False,
+            help="Delete report even if not in 'pending' status")
+
+        options, args = parser.parse_args(args)
+        try:
+            self.report_id, = args
+        except:
+            raise InvalidCommandLine('Specify one report ID')
+
+        self.employee_id = options.employee_id
+        self.force = options.force
+
+    def __call__(self):
+        try:
+            count = delete_report(self.employee_id, self.report_id, self.force)
+        except NoSuchReport:
+            self.submitter.blather("No such report: %s/%s"
+                                   % (self.employee_id, self.report_id))
+        except BadReportStatus as e:
+            self.submitter.blather("Invalid report status: %s/%s, %s"
+                                   % (self.employee_id, self.report_id,
+                                      str(e)))
+        else:
+            self.submitter.blather("Deleted report: %s/%s"
+                                   % (self.employee_id, self.report_id))
+            self.submitter.blather("Removed %d items." % count)
+
+
+_COMMANDS = {
+    'create': CreateReport,
+    'update': UpdateReport,
+    'delete': DeleteReport,
+}
+
+
+def get_description(command):
+    klass = _COMMANDS[command]
+    doc = getattr(klass, '__doc__', '')
+    if doc is None:
+        return ''
+    return ' '.join([x.lstrip() for x in doc.split('\n')])
+
+
+class SubmitExpenses(object):
+    """ Driver for the :command:`submit_expenses` command-line script.
+    """
+    def __init__(self, argv=None, logger=None):
+        self.commands = []
+        if logger is None:
+            logger = self._print
+        self.logger = logger
+        self.parse_arguments(argv)
+
+    def parse_arguments(self, argv=None):
+        """ Parse subcommands and their options from an argv list.
+        """
+        # Global options (not bound to sub-command)
+        mine = []
+        queue = [(None, mine)]
+
+        def _recordCommand(arg):
+            if arg is not None:
+                queue.append((arg, []))
+
+        for arg in argv:
+            if arg in _COMMANDS:
+                _recordCommand(arg)
+            else:
+                queue[-1][1].append(arg)
+
+        _recordCommand(None)
+
+        usage = ("%prog [GLOBAL_OPTIONS] "
+                 "[command [COMMAND_OPTIONS]* [COMMAND_ARGS]]")
+        parser = optparse.OptionParser(usage=usage)
+
+        parser.add_option(
+            '-s', '--help-commands',
+            action='store_true',
+            dest='help_commands',
+            help="Show command help")
+
+        parser.add_option(
+            '-q', '--quiet',
+            action='store_const', const=0,
+            dest='verbose',
+            help="Run quietly")
+
+        parser.add_option(
+            '-v', '--verbose',
+            action='count',
+            dest='verbose',
+            default=1,
+            help="Increase verbosity")
+
+        options, args = parser.parse_args(mine)
+
+        self.options = options
+
+        for arg in args:
+            self.commands.append(NotACommand(arg))
+            options.help_commands = True
+
+        if options.help_commands:
+            keys = sorted(_COMMANDS.keys())
+            self.error('Valid commands are:')
+            for x in keys:
+                self.error(' %s' % x)
+                doc = get_description(x)
+                if doc:
+                    self.error(textwrap.fill(doc,
+                                             initial_indent='    ',
+                                             subsequent_indent='    '))
+            return
+
+        for command_name, args in queue:
+            if command_name is not None:
+                command = _COMMANDS[command_name](self, *args)
+                self.commands.append(command)
+
+    def __call__(self):
+        """ Invoke sub-commands parsed by :meth:`parse_arguments`.
+        """
+        if not self.commands:
+            raise InvalidCommandLine('No commands specified')
+
+        for command in self.commands:
+            command()
+
+    def _print(self, text):  # pragma NO COVERAGE
+        sys.stdout.write('%s\n' % text)
+
+    def error(self, text):
+        self.logger(text)
+
+    def blather(self, text, min_level=1):
+        if self.options.verbose >= min_level:
+            self.logger(text)
+
+
+def main(argv=sys.argv[1:]):
+    try:
+        SubmitExpenses(argv)()
+    except InvalidCommandLine as e:  # pragma NO COVERAGE
+        sys.stdout.write('%s\n' % str(e))
+        sys.exit(1)
diff --git a/docs/examples/expenses/setup.py b/docs/examples/expenses/setup.py
new file mode 100644
index 000000000000..c8a7bcb43cee
--- /dev/null
+++ b/docs/examples/expenses/setup.py
@@ -0,0 +1,19 @@
+from setuptools import find_packages
+from setuptools import setup
+
+setup(
+    name='expenses',
+    version='0.1',
+    description='Expense report upload (example for gcloud.datastore)',
+    packages=find_packages(),
+    include_package_data=True,
+    install_requires=[
+        'gcloud',
+    ],
+    entry_points={
+        'console_scripts': [
+            'submit_expenses = expenses.scripts.submit_expenses:main',
+            'review_expenses = expenses.scripts.review_expenses:main',
+        ],
+    },
+)
diff --git a/docs/index.rst b/docs/index.rst
index d65017fe6636..f7a8ab129a34 100644
--- a/docs/index.rst
+++ b/docs/index.rst
@@ -7,6 +7,7 @@
   datastore-api
   datastore-getting-started
   datastore-quickstart
+  datastore-narrative_example
   getting-started
   storage-api
   storage-getting-started
diff --git a/run_pylint.py b/run_pylint.py
index 21560f318112..55159f6f5af5 100644
--- a/run_pylint.py
+++ b/run_pylint.py
@@ -32,8 +32,10 @@ def is_production_filename(filename):
     :rtype: `bool`
     :returns: Boolean indicating production status.
     """
-    return not ('demo' in filename or 'test' in filename
-                or filename.startswith('regression'))
+    return not ('demo' in filename or
+                'test' in filename or
+                'docs' in filename or
+                filename.startswith('regression'))
 
 
 def get_python_files():