diff --git a/rbql/rbql_ipython.py b/rbql/rbql_ipython.py
new file mode 100755
index 0000000..e58bc77
--- /dev/null
+++ b/rbql/rbql_ipython.py
@@ -0,0 +1,117 @@
+# -*- coding: utf-8 -*-
+from __future__ import unicode_literals
+from __future__ import print_function
+
+from . import rbql_engine
+from . import rbql_pandas
+
+# TODO figure out how to implement at least basic autocomplete for the magic command.
+
+import re
+from_autocomplete_matcher = re.compile(r'(?:^| )from +([_a-zA-Z0-9]+)(?:$| )', flags=re.IGNORECASE)
+join_autocomplete_matcher = re.compile(r'(?:^| )join +([_a-zA-Z0-9]+)(?:$| )', flags=re.IGNORECASE)
+
+
+class IPythonDataframeRegistry(rbql_engine.RBQLTableRegistry):
+    # TODO consider making this class nested under load_ipython_extension to avoid redundant `import pandas`.
+    def __init__(self, all_ns_refs):
+        self.all_ns_refs = all_ns_refs
+
+    def get_iterator_by_table_id(self, table_id, single_char_alias):
+        import pandas
+        # It seems to be the first namespace is "user" namespace, at least according to this code: 
+        # https://github.com/google/picatrix/blob/a2f39766ad4b007b125dc8f84916e18fb3dc5478/picatrix/lib/utils.py
+        for ns in self.all_ns_refs:
+            if table_id in ns and isinstance(ns[table_id], pandas.DataFrame):
+                return rbql_pandas.DataframeIterator(ns[table_id], normalize_column_names=True, variable_prefix=single_char_alias)
+        return None
+
+
+def eprint(*args, **kwargs):
+    import sys
+    print(*args, file=sys.stderr, **kwargs)
+
+
+class AttrDict(dict):
+    # Helper class to convert dict keys to attributes. See explanation here: https://stackoverflow.com/a/14620633/2898283
+    def __init__(self, *args, **kwargs):
+        super(AttrDict, self).__init__(*args, **kwargs)
+        self.__dict__ = self
+
+
+def load_ipython_extension(ipython):
+    from IPython.core.magic import register_line_magic
+    from IPython.core.getipython import get_ipython
+    import pandas
+
+    ipython = ipython or get_ipython() # The pattern taken from here: https://github.com/pydoit/doit/blob/9efe141a5dc96d4912143561695af7fc4a076490/doit/tools.py
+    # ipython is interactiveshell. Docs: https://ipython.readthedocs.io/en/stable/api/generated/IPython.core.interactiveshell.html
+
+
+    def get_table_column_names(table_id):
+        user_namespace = ipython.all_ns_refs[0] if len(ipython.all_ns_refs) else dict()
+        if table_id not in user_namespace or not isinstance(user_namespace[table_id], pandas.DataFrame):
+            return []
+        input_df = user_namespace[table_id]
+        if isinstance(input_df.columns, pandas.RangeIndex) or not len(input_df.columns):
+            return []
+        return [str(v) for v in list(input_df.columns)]
+
+
+    def rbql_completers(self, event):
+        # This should return a list of strings with possible completions.
+        # Note that all the included strings that don't start with event.symbol
+        # are removed, in order to not confuse readline.
+
+        # eg Typing %%rbql foo then hitting tab would yield an event like so: namespace(command='%%rbql', line='%%rbql foo', symbol='foo', text_until_cursor='%%rbql foo')
+        # https://stackoverflow.com/questions/36479197/ipython-custom-tab-completion-for-user-magic-function
+        # https://github.com/ipython/ipython/issues/11878
+
+        simple_sql_keys_lower_case = ['update', 'select', 'where', 'limit', 'from', 'group by', 'order by']
+        simple_sql_keys_upper_case = [sk.upper() for sk in simple_sql_keys_lower_case]
+        autocomplete_suggestions = simple_sql_keys_lower_case + simple_sql_keys_upper_case
+
+        if event.symbol and event.symbol.startswith('a.'):
+            from_match = from_autocomplete_matcher.search(event.line)
+            if from_match is not None:
+                table_id = from_match.group(1)
+                table_column_names = get_table_column_names(table_id)
+                autocomplete_suggestions += ['a.' + cn for cn in table_column_names]
+
+        if event.symbol and event.symbol.startswith('b.'):
+            from_match = join_autocomplete_matcher.search(event.line)
+            if from_match is not None:
+                table_id = from_match.group(1)
+                table_column_names = get_table_column_names(table_id)
+                autocomplete_suggestions += ['b.' + cn for cn in table_column_names]
+        
+        return autocomplete_suggestions
+
+    ipython.set_hook('complete_command', rbql_completers, str_key='%rbql')
+
+
+    # The difference between line and cell magic is described here: https://jakevdp.github.io/PythonDataScienceHandbook/01.03-magic-commands.html.
+    # In short: line magic only accepts one line of input whereas cell magic supports multiline input as magic command argument.
+    # Both line and cell magic would make sense for RBQL queries but for MVP it should be enough to implement just the cell magic.
+    @register_line_magic("rbql")
+    def run_rbql_query(query_text):
+        # Unfortunately globals() and locals() called from here won't contain user variables defined in the notebook.
+
+        tables_registry = IPythonDataframeRegistry(ipython.all_ns_refs)
+        output_writer = rbql_pandas.DataframeWriter()
+        # Ignore warnings because pandas dataframes can't cause them.
+        output_warnings = []
+        # TODO make it possible to specify user_init_code in code cells.
+        error_type, error_msg = None, None
+        user_namespace = None
+        if len(ipython.all_ns_refs) > 0:
+            user_namespace = AttrDict(ipython.all_ns_refs[0])
+        try:
+            rbql_engine.query(query_text, input_iterator=None, output_writer=output_writer, output_warnings=output_warnings, join_tables_registry=tables_registry, user_init_code='', user_namespace=user_namespace)
+        except Exception as e:
+            error_type, error_msg = rbql_engine.exception_to_error_info(e)
+        if error_type is None:
+            return output_writer.result
+        else:
+            # TODO use IPython.display to print error in red color, see https://stackoverflow.com/questions/16816013/is-it-possible-to-print-using-different-colors-in-ipythons-notebook
+            eprint('Error [{}]: {}'.format(error_type, error_msg))
diff --git a/rbql/rbql_pandas.py b/rbql/rbql_pandas.py
new file mode 100755
index 0000000..0abd0d1
--- /dev/null
+++ b/rbql/rbql_pandas.py
@@ -0,0 +1,96 @@
+# -*- coding: utf-8 -*-
+from __future__ import unicode_literals
+from __future__ import print_function
+
+from . import rbql_engine
+
+
+def get_dataframe_column_names_for_rbql(dataframe):
+    import pandas
+    if isinstance(dataframe.columns, pandas.RangeIndex) or not len(dataframe.columns):
+        return None
+    return [str(v) for v in list(dataframe.columns)]
+
+
+class DataframeIterator(rbql_engine.RBQLInputIterator):
+    def __init__(self, table, normalize_column_names=True, variable_prefix='a'):
+        self.table = table
+        self.normalize_column_names = normalize_column_names
+        self.variable_prefix = variable_prefix
+        self.NR = 0
+        # TODO include `Index` into the list of addressable variable names.
+        self.column_names = get_dataframe_column_names_for_rbql(table)
+        self.table_itertuples = self.table.itertuples(index=False)
+
+    def get_variables_map(self, query_text):
+        variable_map = dict()
+        rbql_engine.parse_basic_variables(query_text, self.variable_prefix, variable_map)
+        rbql_engine.parse_array_variables(query_text, self.variable_prefix, variable_map)
+        if self.column_names is not None:
+            if self.normalize_column_names:
+                rbql_engine.parse_dictionary_variables(query_text, self.variable_prefix, self.column_names, variable_map)
+                rbql_engine.parse_attribute_variables(query_text, self.variable_prefix, self.column_names, 'column names list', variable_map)
+            else:
+                rbql_engine.map_variables_directly(query_text, self.column_names, variable_map)
+        return variable_map
+
+    def get_record(self):
+        try:
+            record = next(self.table_itertuples)
+        except StopIteration:
+            return None
+        self.NR += 1
+        # Convert to list because `record` has `Pandas` type.
+        return list(record)
+
+    def get_warnings(self):
+        return []
+
+    def get_header(self):
+        return self.column_names
+
+
+class DataframeWriter(rbql_engine.RBQLOutputWriter):
+    def __init__(self):
+        self.header = None
+        self.output_rows = []
+        self.result = None
+
+    def write(self, fields):
+        self.output_rows.append(fields)
+        return True
+
+    def set_header(self, header):
+        self.header = header
+
+    def finish(self):
+        import pandas as pd
+        self.result = pd.DataFrame(self.output_rows, columns=self.header)
+
+
+class SingleDataframeRegistry(rbql_engine.RBQLTableRegistry):
+    def __init__(self, table, table_name, normalize_column_names=True):
+        self.table = table
+        self.normalize_column_names = normalize_column_names
+        self.table_name = table_name
+
+    def get_iterator_by_table_id(self, table_id, single_char_alias):
+        if table_id.lower() != self.table_name:
+            raise rbql_engine.RbqlParsingError('Unable to find join table: "{}"'.format(table_id))
+        return DataframeIterator(self.table, self.normalize_column_names, single_char_alias)
+
+
+def query_dataframe(query_text, input_dataframe, output_warnings=None, join_dataframe=None, normalize_column_names=True, user_init_code=''):
+    if output_warnings is None:
+        # Ignore output warnings if the output_warnings container hasn't been provided.
+        output_warnings = []
+    if not normalize_column_names and join_dataframe is not None:
+        input_columns = get_dataframe_column_names_for_rbql(input_dataframe)
+        join_columns = get_dataframe_column_names_for_rbql(join_dataframe)
+        if input_columns is not None and join_columns is not None:
+            rbql_engine.ensure_no_ambiguous_variables(query_text, input_columns, join_columns)
+    input_iterator = DataframeIterator(input_dataframe, normalize_column_names)
+    output_writer = DataframeWriter()
+    join_tables_registry = None if join_dataframe is None else SingleDataframeRegistry(join_dataframe, 'b', normalize_column_names)
+    rbql_engine.query(query_text, input_iterator, output_writer, output_warnings, join_tables_registry, user_init_code=user_init_code)
+    return output_writer.result
diff --git a/rbql/rbql_sqlite.py b/rbql/rbql_sqlite.py
new file mode 100755
index 0000000..3ad0feb
--- /dev/null
+++ b/rbql/rbql_sqlite.py
@@ -0,0 +1,105 @@
+# -*- coding: utf-8 -*-
+
+# This module allows to query sqlite databases using RBQL
+
+from __future__ import unicode_literals
+from __future__ import print_function
+
+
+# TODO consider to support table names in "FROM" section of the query, making table_name param of SqliteRecordIterator optional
+# TODO consider adding support for multiple variable_prefixes i.e. "a" and <table_name> or "b" and <join_table_name> to alias input and join tables
+
+
+import re
+import os
+import sys
+
+from . import rbql_engine
+from . import rbql_csv
+
+
+class SqliteRecordIterator(rbql_engine.RBQLInputIterator):
+    def __init__(self, db_connection, table_name, variable_prefix='a'):
+        self.db_connection = db_connection
+        self.table_name = table_name
+        self.variable_prefix = variable_prefix
+        self.cursor = self.db_connection.cursor()
+        import sqlite3
+        if re.match('^[a-zA-Z0-9_]*$', table_name) is None:
+            raise rbql_engine.RbqlIOHandlingError('Unable to use "{}": input table name can contain only alphanumeric characters and underscore'.format(table_name))
+        try:
+            self.cursor.execute('SELECT * FROM {};'.format(table_name))
+        except sqlite3.OperationalError as e:
+            if str(e).find('no such table') != -1:
+                raise rbql_engine.RbqlIOHandlingError('no such table "{}"'.format(table_name))
+            raise
+
+    def get_header(self):
+        column_names = [description[0] for description in self.cursor.description]
+        return column_names
+
+    def get_variables_map(self, query_text):
+        variable_map = dict()
+        rbql_engine.parse_basic_variables(query_text, self.variable_prefix, variable_map)
+        rbql_engine.parse_array_variables(query_text, self.variable_prefix, variable_map)
+        rbql_engine.parse_dictionary_variables(query_text, self.variable_prefix, self.get_header(), variable_map)
+        rbql_engine.parse_attribute_variables(query_text, self.variable_prefix, self.get_header(), 'table column names', variable_map)
+        return variable_map
+
+    def get_record(self):
+        record_tuple = self.cursor.fetchone()
+        if record_tuple is None:
+            return None
+        # We need to convert tuple to list here because otherwise we won't be able to concatinate lists in expressions with star `*` operator
+        return list(record_tuple)
+
+    def get_all_records(self, num_rows=None):
+        # TODO consider to use TOP in the sqlite query when num_rows is not None
+        if num_rows is None:
+            return self.cursor.fetchall()
+        result = []
+        for i in range(num_rows):
+            row = self.cursor.fetchone()
+            if row is None:
+                break
+            result.append(row)
+        return result
+
+    def get_warnings(self):
+        return []
+
+
+class SqliteDbRegistry(rbql_engine.RBQLTableRegistry):
+    def __init__(self, db_connection):
+        self.db_connection = db_connection
+
+    def get_iterator_by_table_id(self, table_id, single_char_alias):
+        self.record_iterator = SqliteRecordIterator(self.db_connection, table_id, single_char_alias)
+        return self.record_iterator
+
+
+def query_sqlite_to_csv(query_text, db_connection, input_table_name, output_path, output_delim, output_policy, output_csv_encoding, output_warnings, user_init_code='', colorize_output=False):
+    output_stream, close_output_on_finish = (None, False)
+    join_tables_registry = None
+    try:
+        output_stream, close_output_on_finish = (sys.stdout, False) if output_path is None else (open(output_path, 'wb'), True)
+
+        if not rbql_csv.is_ascii(query_text) and output_csv_encoding == 'latin-1':
+            raise rbql_engine.RbqlIOHandlingError('To use non-ascii characters in query enable UTF-8 encoding instead of latin-1/binary')
+
+        if not rbql_csv.is_ascii(output_delim) and output_csv_encoding == 'latin-1':
+            raise rbql_engine.RbqlIOHandlingError('To use non-ascii separators enable UTF-8 encoding instead of latin-1/binary')
+
+        default_init_source_path = os.path.join(os.path.expanduser('~'), '.rbql_init_source.py')
+        if user_init_code == '' and os.path.exists(default_init_source_path):
+            user_init_code = rbql_csv.read_user_init_code(default_init_source_path)
+
+        join_tables_registry = SqliteDbRegistry(db_connection)
+        input_iterator = SqliteRecordIterator(db_connection, input_table_name)
+        output_writer = rbql_csv.CSVWriter(output_stream, close_output_on_finish, output_csv_encoding, output_delim, output_policy, colorize_output=colorize_output)
+        rbql_engine.query(query_text, input_iterator, output_writer, output_warnings, join_tables_registry, user_init_code)
+    finally:
+        if close_output_on_finish:
+            output_stream.close()
+
+