diff --git a/app.py b/app.py
index 39e7a7d..24f0193 100644
--- a/app.py
+++ b/app.py
@@ -7,10 +7,12 @@ import json
 from pytz import timezone
 import requests
 
+import click
 from flask import Flask
+from flask.cli import AppGroup
 from flask_migrate import Migrate
 from flask_sqlalchemy import SQLAlchemy
-from sqlalchemy import Boolean, Column, Date, Integer, String, Time
+from sqlalchemy import Boolean, Column, Date, DateTime, Integer, String, Time
 from tqdm import tqdm
 
 import config
@@ -18,6 +20,10 @@ import config
 
 app = Flask(__name__)
 
+cli = AppGroup('tgvmax', help="Manage the TGVMax dataset.")
+app.cli.add_command(cli)
+
+
 app.config |= config.FLASK_CONFIG
 
 db = SQLAlchemy(app)
@@ -29,19 +35,25 @@ class Train(db.Model):
     id = Column(String, primary_key=True)
     day = Column(Date, index=True)
     number = Column(Integer, index=True)
-    entity = Column(String(255))
-    axe = Column(String(255), index=True)
+    entity = Column(String(10))
+    axe = Column(String(32), index=True)
     orig_iata = Column(String(5), index=True)
     dest_iata = Column(String(5), index=True)
-    orig = Column(String(255))
-    dest = Column(String(255))
-    dep = Column(String(255))
+    orig = Column(String(32))
+    dest = Column(String(32))
+    dep = Column(Time)
     arr = Column(Time)
     tgvmax = Column(Boolean, index=True)
-    remaining_seats = Column(Integer)
+    remaining_seats = Column(Integer, default=-1)
+    last_modification = Column(DateTime)
+    expiration_time = Column(DateTime)
 
 
+@cli.command("update-dataset")
 def update_dataset():
+    """
+    Query the latest version of the SNCF OpenData dataset, as a CSV file.
+    """
     try:
         resp = requests.get('https://ressources.data.sncf.com/explore/dataset/tgvmax/information/')
         content = resp.content.decode().split('<script type="application/ld+json">')[1].split('</script>')[0].strip()
@@ -76,37 +88,61 @@ def update_dataset():
         print(e)
 
 
-def parse_trains(*, filter_day: date | None = None,
-                 filter_number: int | None = None,
-                 filter_tgvmax: bool | None = None):
-    trains = []
+@cli.command("parse-csv")
+@click.option('-F', '--flush', type=bool, is_flag=True, help="Flush the database before filling it.")
+def parse_trains(flush: bool = False):
+    """
+    Parse the CSV file and store it to the database.
+    """
+
+    if flush:
+        print("Flush database…")
+        db.session.query(Train).delete()
+
+    last_modification = datetime.utcfromtimestamp(os.path.getmtime('tgvmax.csv')).replace(tzinfo=timezone('UTC'))
 
     with open('tgvmax.csv') as f:
         first_line = True
-        for line in csv.reader(f, delimiter=';'):
+        already_seen = set()
+        for line in tqdm(csv.reader(f, delimiter=';')):
             if first_line:
                 first_line = False
                 continue
 
-            train = Train(*line)
-            train.day = date.fromisoformat(train.day)
-            train.number = int(train.number)
-            train.dep = time.fromisoformat(train.dep)
-            train.arr = time.fromisoformat(train.arr)
-            train.tgvmax = train.tgvmax == 'OUI'
-
-            if filter_day is not None and train.day != filter_day:
+            train_id = f"{line[1]}-{line[0]}-{line[4]}-{line[5]}"
+            if train_id in already_seen:
+                # Some trains are mysteriously duplicated, concerns only some « Intercités de nuit »
+                # and the Brive-la-Gaillarde -- Paris
+                # and, maybe, for Roubaix-Tourcoing
+                if line[3] != "IC NUIT" and line[1] != '3614' and not (line[4] == 'FRADP' and line[5] == 'FRADM'):
+                    print("Duplicate:", train_id)
                 continue
 
-            if filter_number is not None and train.number != filter_number:
-                continue
+            train = Train(
+                id=train_id,
+                day=date.fromisoformat(line[0]),
+                number=int(line[1]),
+                entity=line[2],
+                axe=line[3],
+                orig_iata=line[4],
+                dest_iata=line[5],
+                orig=line[6],
+                dest=line[7],
+                dep=time.fromisoformat(line[8]),
+                arr=time.fromisoformat(line[9]),
+                tgvmax=line[10] == 'OUI',
+                last_modification=last_modification,
+                expiration_time=last_modification,
+            )
+            if flush:
+                db.session.add(train)
+            else:
+                db.session.merge(train)
 
-            if filter_tgvmax is not None and train.tgvmax != filter_tgvmax:
-                continue
+            if line[3] == "IC NUIT" or line[1] == '3614' or (line[4] == 'FRADP' and line[5] == 'FRADM'):
+                already_seen.add(train_id)
 
-            trains.append(train)
-
-    return trains
+    db.session.commit()
 
 
 def find_routes(day, orig, dest):