diff --git a/.gitignore b/.gitignore
index 25a68ea..7229501 100644
--- a/.gitignore
+++ b/.gitignore
@@ -2,3 +2,4 @@ __pycache__/
 venv/
 .coverage
 client/build/
+*.pyc
diff --git a/client/requirements.txt b/client/requirements.txt
new file mode 100644
index 0000000..d6e9c5d
--- /dev/null
+++ b/client/requirements.txt
@@ -0,0 +1,17 @@
+attrs==19.3.0
+coverage==5.1
+importlib-metadata==1.6.0
+more-itertools==8.2.0
+numpy==1.18.4
+packaging==20.3
+pandas==1.0.3
+pluggy==0.13.1
+py==1.8.1
+pyparsing==2.4.7
+pytest==5.4.1
+python-dateutil==2.8.1
+pytz==2020.1
+six==1.14.0
+wcwidth==0.1.9
+yapf==0.30.0
+zipp==3.1.0
diff --git a/client/streamsql/__init__.py b/client/streamsql/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/client/streamsql/errors.py b/client/streamsql/errors.py
new file mode 100644
index 0000000..eefe394
--- /dev/null
+++ b/client/streamsql/errors.py
@@ -0,0 +1,8 @@
+class Error(Exception):
+    """Base class for public StreamSQL client errors."""
+
+
+class TableAlreadyExists(Error):
+    """Raised when the requested table name already exists."""
+    def __init__(self, name):
+        self.message = "Table with name \"{0}\" already exists".format(name)
diff --git a/client/streamsql/local.py b/client/streamsql/local.py
new file mode 100644
index 0000000..82209fa
--- /dev/null
+++ b/client/streamsql/local.py
@@ -0,0 +1,83 @@
+from streamsql.errors import TableAlreadyExists
+import pandas
+from pandasql import sqldf
+
+
+class FeatureStore:
+    """Feature is a local in-memory feature store
+
+    The local feature store provides a subset of the core functionality
+    of the backend one. It should only be used for models that are
+    trained and used locally.
+    """
+    def __init__(self):
+        self._tables = dict()
+
+    def create_table_from_csv(self, csv_file, table_name="", primary_key=""):
+        """Create a table from a local csv file"""
+        if table_name in self._tables:
+            raise TableAlreadyExists(table_name)
+
+        table = Table.from_csv(csv_file=csv_file, primary_key=primary_key)
+        self._tables[table_name] = table
+        return table
+
+    def get_table(self, table_name):
+        return self._tables[table_name]
+
+    def has_table(self, table_name):
+        return table_name in self._tables
+
+    def materialize_table(
+        self,
+        name="",
+        query="",
+        dependencies=[],
+        output_columns=[],
+        primary_key="",
+    ):
+        """Create a Table by applying a SQL query on other Tables"""
+        query_ctx = {
+            dep: self.get_table(dep)._dataframe
+            for dep in dependencies
+        }
+        dataframe = sqldf(query, query_ctx)
+        dataframe.columns = output_columns
+        dataframe.set_index(keys=primary_key,
+                            inplace=True,
+                            verify_integrity=True)
+        table = Table(dataframe)
+        self._tables[name] = table
+        return table
+
+
+class Table:
+    """Table is an in-memory implementation of the StreamSQL table"""
+    @classmethod
+    def from_csv(cls, csv_file="", primary_key=""):
+        """Create a Table from a CSV file."""
+        dataframe = Table._dataframe_from_csv(csv_file, primary_key)
+        return cls(dataframe)
+
+    def __init__(self, dataframe):
+        """Create a Table from a pandas.DataFrame"""
+        self._dataframe = dataframe
+
+    def lookup(self, key):
+        """Lookup returns an array from a table by its primary key"""
+        item_series = self._dataframe.loc[key]
+        return item_series.to_list()
+
+    def __eq__(self, other):
+        if isinstance(other, Table):
+            return self._dataframe.equals(other._dataframe)
+        else:
+            return NotImplemented
+
+    def _dataframe_from_csv(file_name, index_col):
+        dataframe = pandas.read_csv(file_name, index_col=index_col)
+        Table._clean_dataframe(dataframe)
+        return dataframe
+
+    def _clean_dataframe(dataframe):
+        dataframe.index = dataframe.index.astype(str, copy=False)
diff --git a/client/tests/test_local_backend.py b/client/tests/test_local_backend.py
new file mode 100644
index 0000000..babb6c1
--- /dev/null
+++ b/client/tests/test_local_backend.py
@@ -0,0 +1,121 @@
+import pytest
+import streamsql.local
+import streamsql.errors
+import os, sys
+import pandas as pd
+
+test_dir = os.path.dirname(os.path.realpath(__file__))
+testdata_dir = os.path.join(test_dir, 'testdata')
+users_file = os.path.join(testdata_dir, 'users.csv')
+purchases_file = os.path.join(testdata_dir, 'purchases.csv')
+
+
+@pytest.fixture
+def feature_store():
+    return streamsql.local.FeatureStore()
+
+
+@pytest.fixture
+def user_table(feature_store):
+    return create_user_table(feature_store)
+
+
+@pytest.fixture
+def purchase_table(feature_store):
+    return create_purchase_table(feature_store)
+
+
+def create_user_table(feature_store):
+    return feature_store.create_table_from_csv(users_file,
+                                               table_name="users",
+                                               primary_key="id")
+
+
+def create_purchase_table(feature_store):
+    return feature_store.create_table_from_csv(purchases_file,
+                                               table_name="purchases",
+                                               primary_key="id")
+
+
+def test_table_already_exists(feature_store):
+    create_user_table(feature_store)
+    with pytest.raises(streamsql.errors.TableAlreadyExists):
+        create_user_table(feature_store)
+
+
+def test_has_table(feature_store):
+    create_user_table(feature_store)
+    assert feature_store.has_table("users")
+
+
+def test_not_has_table(feature_store):
+    create_user_table(feature_store)
+    assert not feature_store.has_table("users2")
+
+
+def test_get_table(feature_store):
+    created_table = create_user_table(feature_store)
+    got_table = feature_store.get_table("users")
+    assert created_table == got_table
+
+
+def test_get_table_fail(feature_store):
+    with pytest.raises(KeyError):
+        feature_store.get_table("users2")
+
+
+def test_table_lookup(user_table):
+    assert user_table.lookup("1") == ["simba"]
+
+
+def test_table_lookup_fail(user_table):
+    with pytest.raises(KeyError):
+        user_table.lookup("abc")
+
+
+def test_table_simple_sql(feature_store):
+    create_user_table(feature_store)
+    nora_table = feature_store.materialize_table(
+        name="nora",
+        query="SELECT id, name FROM users WHERE name == 'nora'",
+        dependencies=["users"],
+        output_columns=["new_id", "new_name"],
+        primary_key=["new_id"],
+    )
+    df = pd.DataFrame(["nora"], index=["2"], columns=["new_name"])
+    df.index.name = "new_id"
+    expected = streamsql.local.Table(df)
+    assert nora_table == expected
+
+
+def test_table_join_sql(feature_store):
+    create_user_table(feature_store)
+    create_purchase_table(feature_store)
+    dollars_spent_table = feature_store.materialize_table(
+        name="dollars_spent",
+        query="""SELECT user.id, SUM(price) FROM purchases purchase
+        INNER JOIN users user ON purchase.user=user.id GROUP BY user.id
+        ORDER BY user.id ASC
+        """,
+        dependencies=["users", "purchases"],
+        output_columns=["user", "spent"],
+        primary_key=["user"],
+    )
+    df = pd.DataFrame([1000, 10], index=["1", "3"], columns=["spent"])
+    df.index.name = "user"
+    expected = streamsql.local.Table(df)
+    assert dollars_spent_table == expected
+
+
+def test_materialized_table_is_stored(feature_store):
+    create_user_table(feature_store)
+    created = feature_store.materialize_table(
+        name="user_cpy",
+        query="SELECT id, name FROM users",
+        dependencies=["users"],
+        output_columns=["id", "name"],
+        primary_key=["id"],
+    )
+    assert feature_store.has_table("user_cpy")
+    got = feature_store.get_table("user_cpy")
+    assert created == got
diff --git a/client/tests/testdata/purchases.csv b/client/tests/testdata/purchases.csv
new file mode 100644
index 0000000..3a6452e
--- /dev/null
+++ b/client/tests/testdata/purchases.csv
@@ -0,0 +1,4 @@
+id,user,item,price
+0,1,laptop,1000
+1,3,peanut butter,5
+2,3,peanut butter,5
diff --git a/client/tests/testdata/users.csv b/client/tests/testdata/users.csv
new file mode 100644
index 0000000..9a31522
--- /dev/null
+++ b/client/tests/testdata/users.csv
@@ -0,0 +1,4 @@
+id,name
+1,simba
+2,nora
+3,chups