sqlite_utils Python library#

Getting started#

Here’s how to create a new SQLite database file containing a new chickens table, populated with four records:

from sqlite_utils import Database

db = Database("chickens.db")
    "name": "Azi",
    "color": "blue",
}, {
    "name": "Lila",
    "color": "blue",
}, {
    "name": "Suna",
    "color": "gold",
}, {
    "name": "Cardi",
    "color": "black",

You can loop through those rows like this:

for row in db["chickens"].rows:

Which outputs the following:

{'name': 'Azi', 'color': 'blue'}
{'name': 'Lila', 'color': 'blue'}
{'name': 'Suna', 'color': 'gold'}
{'name': 'Cardi', 'color': 'black'}

To run a SQL query, use db.query():

for row in db.query("""
    select color, count(*)
    from chickens group by color
    order by count(*) desc

Which outputs:

{'color': 'blue', 'count(*)': 2}
{'color': 'gold', 'count(*)': 1}
{'color': 'black', 'count(*)': 1}

Connecting to or creating a database#

Database objects are constructed by passing in either a path to a file on disk or an existing SQLite3 database connection:

from sqlite_utils import Database

db = Database("my_database.db")

This will create my_database.db if it does not already exist.

If you want to recreate a database from scratch (first removing the existing file from disk if it already exists) you can use the recreate=True argument:

db = Database("my_database.db", recreate=True)

Instead of a file path you can pass in an existing SQLite connection:

import sqlite3

db = Database(sqlite3.connect("my_database.db"))

If you want to create an in-memory database, you can do so like this:

db = Database(memory=True)

You can also create a named in-memory database. Unlike regular memory databases these can be accessed by multiple threads, provided at least one reference to the database still exists. del db will clear the database from memory.

db = Database(memory_name="my_shared_database")

Connections use PRAGMA recursive_triggers=on by default. If you don’t want to use recursive triggers you can turn them off using:

db = Database(memory=True, recursive_triggers=False)

By default, any sqlite-utils plugins that implement the prepare_connection(conn) hook will be executed against the connection when you create the Database object. You can opt out of executing plugins using execute_plugins=False like this:

db = Database(memory=True, execute_plugins=False)

You can pass strict=True to enable SQLite STRICT mode for all tables created using this database object:

db = Database("my_database.db", strict=True)

Attaching additional databases#

SQLite supports cross-database SQL queries, which can join data from tables in more than one database file.

You can attach an additional database using the .attach() method, providing an alias to use for that database and the path to the SQLite file on disk.

db = Database("first.db")
db.attach("second", "second.db")
# Now you can run queries like this one:
select * from table_in_first
    union all
select * from second.table_in_second

You can reference tables in the attached database using the alias value you passed to db.attach(alias, filepath) as a prefix, for example the second.table_in_second reference in the SQL query above.

Tracing queries#

You can use the tracer mechanism to see SQL queries that are being executed by SQLite. A tracer is a function that you provide which will be called with sql and params arguments every time SQL is executed, for example:

def tracer(sql, params):
    print("SQL: {} - params: {}".format(sql, params))

You can pass this function to the Database() constructor like so:

db = Database(memory=True, tracer=tracer)

You can also turn on a tracer function temporarily for a block of code using the with db.tracer(...) context manager:

db = Database(memory=True)
# ... later
with db.tracer(print):
    db["dogs"].insert({"name": "Cleo"})

This example will print queries only for the duration of the with block.

Executing queries#

The Database class offers several methods for directly executing SQL queries.

db.query(sql, params)#

The db.query(sql) function executes a SQL query and returns an iterator over Python dictionaries representing the resulting rows:

db = Database(memory=True)
db["dogs"].insert_all([{"name": "Cleo"}, {"name": "Pancakes"}])
for row in db.query("select * from dogs"):
# Outputs:
# {'name': 'Cleo'}
# {'name': 'Pancakes'}

db.execute(sql, params)#

The db.execute() and db.executescript() methods provide wrappers around .execute() and .executescript() on the underlying SQLite connection. These wrappers log to the tracer function if one has been registered.

db.execute(sql) returns a sqlite3.Cursor that was used to execute the SQL.

db = Database(memory=True)
db["dogs"].insert({"name": "Cleo"})
cursor = db.execute("update dogs set name = 'Cleopaws'")
# Outputs the number of rows affected by the update
# In this case 2

Other cursor methods such as .fetchone() and .fetchall() are also available, see the standard library documentation.

Passing parameters#

Both db.query() and db.execute() accept an optional second argument for parameters to be passed to the SQL query.

This can take the form of either a tuple/list or a dictionary, depending on the type of parameters used in the query. Values passed in this way will be correctly quoted and escaped, helping avoid SQL injection vulnerabilities.

? parameters in the SQL query can be filled in using a list:

db.execute("update dogs set name = ?", ["Cleopaws"])
# This will rename ALL dogs to be called "Cleopaws"

Named parameters using :name can be filled using a dictionary:

dog = next(db.query(
    "select rowid, name from dogs where name = :name",
    {"name": "Cleopaws"}
# dog is now {'rowid': 1, 'name': 'Cleopaws'}

In this example next() is used to retrieve the first result in the iterator returned by the db.query() method.

Accessing tables#

Tables are accessed using the indexing operator, like so:

table = db["my_table"]

If the table does not yet exist, it will be created the first time you attempt to insert or upsert data into it.

You can also access tables using the .table() method like so:

table = db.table("my_table")

Using this factory function allows you to set Table configuration options.

Listing tables#

You can list the names of tables in a database using the .table_names() method:

>>> db.table_names()

To see just the FTS4 tables, use .table_names(fts4=True). For FTS5, use .table_names(fts5=True).

You can also iterate through the table objects themselves using the .tables property:

>>> db.tables
[<Table dogs>]

Listing views#

.view_names() shows you a list of views in the database:

>>> db.view_names()

You can iterate through view objects using the .views property:

>>> db.views
[<View good_dogs>]

View objects are similar to Table objects, except that any attempts to insert or update data will throw an error. The full list of methods and properties available on a view object is as follows:

  • columns

  • columns_dict

  • count

  • schema

  • rows

  • rows_where(where, where_args, order_by, select)

  • drop()

Listing rows#

To iterate through dictionaries for each of the rows in a table, use .rows:

>>> db = sqlite_utils.Database("dogs.db")
>>> for row in db["dogs"].rows:
...     print(row)
{'id': 1, 'age': 4, 'name': 'Cleo'}
{'id': 2, 'age': 2, 'name': 'Pancakes'}

You can filter rows by a WHERE clause using .rows_where(where, where_args):

>>> db = sqlite_utils.Database("dogs.db")
>>> for row in db["dogs"].rows_where("age > ?", [3]):
...     print(row)
{'id': 1, 'age': 4, 'name': 'Cleo'}

The first argument is a fragment of SQL. The second, optional argument is values to be passed to that fragment - you can use ? placeholders and pass an array, or you can use :named parameters and pass a dictionary, like this:

>>> for row in db["dogs"].rows_where("age > :age", {"age": 3}):
...     print(row)
{'id': 1, 'age': 4, 'name': 'Cleo'}

To return custom columns (instead of the default that uses select *) pass select="column1, column2":

>>> db = sqlite_utils.Database("dogs.db")
>>> for row in db["dogs"].rows_where(select='name, age'):
...     print(row)
{'name': 'Cleo', 'age': 4}

To specify an order, use the order_by= argument:

>>> for row in db["dogs"].rows_where("age > 1", order_by="age"):
...     print(row)
{'id': 2, 'age': 2, 'name': 'Pancakes'}
{'id': 1, 'age': 4, 'name': 'Cleo'}

You can use order_by="age desc" for descending order.

You can order all records in the table by excluding the where argument:

>>> for row in db["dogs"].rows_where(order_by="age desc"):
...     print(row)
{'id': 1, 'age': 4, 'name': 'Cleo'}
{'id': 2, 'age': 2, 'name': 'Pancakes'}

This method also accepts offset= and limit= arguments, for specifying an OFFSET and a LIMIT for the SQL query:

>>> for row in db["dogs"].rows_where(order_by="age desc", limit=1):
...     print(row)
{'id': 1, 'age': 4, 'name': 'Cleo'}

Counting rows#

To count the number of rows that would be returned by a where filter, use .count_where(where, where_args):

>>> db["dogs"].count_where("age > ?", [1])

Listing rows with their primary keys#

Sometimes it can be useful to retrieve the primary key along with each row, in order to pass that key (or primary key tuple) to the .get() or .update() methods.

The .pks_and_rows_where() method takes the same signature as .rows_where() (with the exception of the select= parameter) but returns a generator that yields pairs of (primary key, row dictionary).

The primary key value will usually be a single value but can also be a tuple if the table has a compound primary key.

If the table is a rowid table (with no explicit primary key column) then that ID will be returned.

>>> db = sqlite_utils.Database(memory=True)
>>> db["dogs"].insert({"name": "Cleo"})
>>> for pk, row in db["dogs"].pks_and_rows_where():
...     print(pk, row)
1 {'rowid': 1, 'name': 'Cleo'}

>>> db["dogs_with_pk"].insert({"id": 5, "name": "Cleo"}, pk="id")
>>> for pk, row in db["dogs_with_pk"].pks_and_rows_where():
...     print(pk, row)
5 {'id': 5, 'name': 'Cleo'}

>>> db["dogs_with_compound_pk"].insert(
...     {"species": "dog", "id": 3, "name": "Cleo"},
...     pk=("species", "id")
... )
>>> for pk, row in db["dogs_with_compound_pk"].pks_and_rows_where():
...     print(pk, row)
('dog', 3) {'species': 'dog', 'id': 3, 'name': 'Cleo'}

Retrieving a specific record#

You can retrieve a record by its primary key using table.get():

>>> db = sqlite_utils.Database("dogs.db")
>>> print(db["dogs"].get(1))
{'id': 1, 'age': 4, 'name': 'Cleo'}

If the table has a compound primary key you can pass in the primary key values as a tuple:

>>> db["compound_dogs"].get(("mixed", 3))

If the record does not exist a NotFoundError will be raised:

from sqlite_utils.db import NotFoundError

    row = db["dogs"].get(5)
except NotFoundError:
    print("Dog not found")

Showing the schema#

The db.schema property returns the full SQL schema for the database as a string:

>>> db = sqlite_utils.Database("dogs.db")
>>> print(db.schema)
    [name] TEXT

Creating tables#

The easiest way to create a new table is to insert a record into it:

from sqlite_utils import Database
import sqlite3

db = Database("dogs.db")
dogs = db["dogs"]
    "name": "Cleo",
    "twitter": "cleopaws",
    "age": 3,
    "is_good_dog": True,

This will automatically create a new table called “dogs” with the following schema:

    name TEXT,
    twitter TEXT,
    age INTEGER,
    is_good_dog INTEGER

You can also specify a primary key by passing the pk= parameter to the .insert() call. This will only be obeyed if the record being inserted causes the table to be created:

    "id": 1,
    "name": "Cleo",
    "twitter": "cleopaws",
    "age": 3,
    "is_good_dog": True,
}, pk="id")

After inserting a row like this, the dogs.last_rowid property will return the SQLite rowid assigned to the most recently inserted record.

The dogs.last_pk property will return the last inserted primary key value, if you specified one. This can be very useful when writing code that creates foreign keys or many-to-many relationships.

Custom column order and column types#

The order of the columns in the table will be derived from the order of the keys in the dictionary, provided you are using Python 3.6 or later.

If you want to explicitly set the order of the columns you can do so using the column_order= parameter:

    "id": 1,
    "name": "Cleo",
    "twitter": "cleopaws",
    "age": 3,
    "is_good_dog": True,
}, pk="id", column_order=("id", "twitter", "name"))

You don’t need to pass all of the columns to the column_order parameter. If you only pass a subset of the columns the remaining columns will be ordered based on the key order of the dictionary.

Column types are detected based on the example data provided. Sometimes you may find you need to over-ride these detected types - to create an integer column for data that was provided as a string for example, or to ensure that a table where the first example was None is created as an INTEGER rather than a TEXT column. You can do this using the columns= parameter:

    "id": 1,
    "name": "Cleo",
    "age": "5",
}, pk="id", columns={"age": int, "weight": float})

This will create a table with the following schema:

    [name] TEXT,
    [age] INTEGER,
    [weight] FLOAT

Explicitly creating a table#

You can directly create a new table without inserting any data into it using the .create() method:

    "id": int,
    "name": str,
    "weight": float,
}, pk="id")

The first argument here is a dictionary specifying the columns you would like to create. Each column is paired with a Python type indicating the type of column. See Adding columns for full details on how these types work.

This method takes optional arguments pk=, column_order=, foreign_keys=, not_null=set() and defaults=dict() - explained below.

A sqlite_utils.utils.sqlite3.OperationalError will be raised if a table of that name already exists.

You can pass ignore=True to ignore that error. You can also use if_not_exists=True to use the SQL CREATE TABLE IF NOT EXISTS pattern to achieve the same effect:

    "id": int,
    "name": str,
}, pk="id", if_not_exists=True)

To drop and replace any existing table of that name, pass replace=True. This is a dangerous operation that will result in loss of existing data in the table.

You can also pass transform=True to have any existing tables transformed to match your new table specification. This is a dangerous operation as it will drop columns that are no longer listed in your call to .create(), so be careful when running this.

    "id": int,
    "name": str,
    "weight": float,
}, pk="id", transform=True)

The transform=True option will update the table schema if any of the following have changed:

  • The specified columns or their types

  • The specified primary key

  • The order of the columns, defined using column_order=

  • The not_null= or defaults= arguments

Changes to foreign_keys= are not currently detected and applied by transform=True.

You can pass strict=True to create a table in STRICT mode:

    "id": int,
    "name": str,
}, strict=True)

Compound primary keys#

If you want to create a table with a compound primary key that spans multiple columns, you can do so by passing a tuple of column names to any of the methods that accept a pk= parameter. For example:

    "id": int,
    "breed": str,
    "name": str,
    "weight": float,
}, pk=("breed", "id"))

This also works for the .insert(), .insert_all(), .upsert() and .upsert_all() methods.

Specifying foreign keys#

Any operation that can create a table (.create(), .insert(), .insert_all(), .upsert() and .upsert_all()) accepts an optional foreign_keys= argument which can be used to set up foreign key constraints for the table that is being created.

If you are using your database with Datasette, Datasette will detect these constraints and use them to generate hyperlinks to associated records.

The foreign_keys argument takes a list that indicates which foreign keys should be created. The list can take several forms. The simplest is a list of columns:


The library will guess which tables you wish to reference based on the column names using the rules described in Adding foreign key constraints.

You can also be more explicit, by passing in a list of tuples:

    ("author_id", "authors", "id")

This means that the author_id column should be a foreign key that references the id column in the authors table.

You can leave off the third item in the tuple to have the referenced column automatically set to the primary key of that table. A full example:

    {"id": 1, "name": "Sally"},
    {"id": 2, "name": "Asheesh"}
], pk="id")
    {"title": "Hedgehogs of the world", "author_id": 1},
    {"title": "How to train your wolf", "author_id": 2},
], foreign_keys=[
    ("author_id", "authors")

Table configuration options#

The .insert(), .upsert(), .insert_all() and .upsert_all() methods each take a number of keyword arguments, some of which influence what happens should they cause a table to be created and some of which affect the behavior of those methods.

You can set default values for these methods by accessing the table through the db.table(...) method (instead of using db["table_name"]), like so:

table = db.table(
    not_null={"name", "score"},
    column_order=("id", "name", "score", "url")
# Now you can call .insert() like so:
table.insert({"id": 1, "name": "Tracy", "score": 5})

The configuration options that can be specified in this way are pk, foreign_keys, column_order, not_null, defaults, batch_size, hash_id, hash_id_columns, alter, ignore, replace, extracts, conversions, columns, strict. These are all documented below.

Setting defaults and not null constraints#

Each of the methods that can cause a table to be created take optional arguments not_null=set() and defaults=dict(). The methods that take these optional arguments are:

  • db.create_table(...)

  • table.create(...)

  • table.insert(...)

  • table.insert_all(...)

  • table.upsert(...)

  • table.upsert_all(...)

You can use not_null= to pass a set of column names that should have a NOT NULL constraint set on them when they are created.

You can use defaults= to pass a dictionary mapping columns to the default value that should be specified in the CREATE TABLE statement.

Here’s an example that uses these features:

    [{"id": 1, "name": "Sally", "score": 2}],
    not_null={"name", "score"},
    defaults={"score": 1},
db["authors"].insert({"name": "Dharma"})

# Outputs:
# [{'id': 1, 'name': 'Sally', 'score': 2},
#  {'id': 3, 'name': 'Dharma', 'score': 1}]
# Outputs:
# CREATE TABLE [authors] (
#     [name] TEXT NOT NULL,
# )

Renaming a table#

The db.rename_table(old_name, new_name) method can be used to rename a table:

db.rename_table("my_table", "new_name_for_my_table")

This executes the following SQL:

ALTER TABLE [my_table] RENAME TO [new_name_for_my_table]

Duplicating tables#

The table.duplicate() method creates a copy of the table, copying both the table schema and all of the rows in that table:


The new authors_copy table will now contain a duplicate copy of the data from authors.

This method raises sqlite_utils.db.NoTable if the table does not exist.

Bulk inserts#

If you have more than one record to insert, the insert_all() method is a much more efficient way of inserting them. Just like insert() it will automatically detect the columns that should be created, but it will inspect the first batch of 100 items to help decide what those column types should be.

Use it like this:

    "id": 1,
    "name": "Cleo",
    "twitter": "cleopaws",
    "age": 3,
    "is_good_dog": True,
}, {
    "id": 2,
    "name": "Marnie",
    "twitter": "MarnieTheDog",
    "age": 16,
    "is_good_dog": True,
}], pk="id", column_order=("id", "twitter", "name"))

The column types used in the CREATE TABLE statement are automatically derived from the types of data in that first batch of rows. Any additional columns in subsequent batches will cause a sqlite3.OperationalError exception to be raised unless the alter=True argument is supplied, in which case the new columns will be created.

The function can accept an iterator or generator of rows and will commit them according to the batch size. The default batch size is 100, but you can specify a different size using the batch_size parameter:

    "id": 1,
    "name": "Name {}".format(i),
} for i in range(10000)), batch_size=1000)

You can skip inserting any records that have a primary key that already exists using ignore=True. This works with both .insert({...}, ignore=True) and .insert_all([...], ignore=True).

You can delete all the existing rows in the table before inserting the new records using truncate=True. This is useful if you want to replace the data in the table.

Pass analyze=True to run ANALYZE against the table after inserting the new records.

Insert-replacing data#

If you try to insert data using a primary key that already exists, the .insert() or .insert_all() method will raise a sqlite3.IntegrityError exception.

This example that catches that exception:

from sqlite_utils.utils import sqlite3

    db["dogs"].insert({"id": 1, "name": "Cleo"}, pk="id")
except sqlite3.IntegrityError:
    print("Record already exists with that primary key")

Importing from sqlite_utils.utils.sqlite3 ensures your code continues to work even if you are using the pysqlite3 library instead of the Python standard library sqlite3 module.

Use the ignore=True parameter to ignore this error:

# This fails silently if a record with id=1 already exists
db["dogs"].insert({"id": 1, "name": "Cleo"}, pk="id", ignore=True)

To replace any existing records that have a matching primary key, use the replace=True parameter to .insert() or .insert_all():

    "id": 1,
    "name": "Cleo",
    "twitter": "cleopaws",
    "age": 3,
    "is_good_dog": True,
}, {
    "id": 2,
    "name": "Marnie",
    "twitter": "MarnieTheDog",
    "age": 16,
    "is_good_dog": True,
}], pk="id", replace=True)


Prior to sqlite-utils 2.0 the .upsert() and .upsert_all() methods worked the same way as .insert(replace=True) does today. See Upserting data for the new behaviour of those methods introduced in 2.0.

Updating a specific record#

You can update a record by its primary key using table.update():

>>> db = sqlite_utils.Database("dogs.db")
>>> print(db["dogs"].get(1))
{'id': 1, 'age': 4, 'name': 'Cleo'}
>>> db["dogs"].update(1, {"age": 5})
>>> print(db["dogs"].get(1))
{'id': 1, 'age': 5, 'name': 'Cleo'}

The first argument to update() is the primary key. This can be a single value, or a tuple if that table has a compound primary key:

>>> db["compound_dogs"].update((5, 3), {"name": "Updated"})

The second argument is a dictionary of columns that should be updated, along with their new values.

You can cause any missing columns to be added automatically using alter=True:

>>> db["dogs"].update(1, {"breed": "Mutt"}, alter=True)

Deleting a specific record#

You can delete a record using table.delete():

>>> db = sqlite_utils.Database("dogs.db")
>>> db["dogs"].delete(1)

The delete() method takes the primary key of the record. This can be a tuple of values if the row has a compound primary key:

>>> db["compound_dogs"].delete((5, 3))

Deleting multiple records#

You can delete all records in a table that match a specific WHERE statement using table.delete_where():

>>> db = sqlite_utils.Database("dogs.db")
>>> # Delete every dog with age less than 3
>>> db["dogs"].delete_where("age < ?", [3])

Calling table.delete_where() with no other arguments will delete every row in the table.

Pass analyze=True to run ANALYZE against the table after deleting the rows.

Upserting data#

Upserting allows you to insert records if they do not exist and update them if they DO exist, based on matching against their primary key.

For example, given the dogs database you could upsert the record for Cleo like so:

    "id": 1,
    "name": "Cleo",
    "twitter": "cleopaws",
    "age": 4,
    "is_good_dog": True,
}, pk="id", column_order=("id", "twitter", "name"))

If a record exists with id=1, it will be updated to match those fields. If it does not exist it will be created.

Any existing columns that are not referenced in the dictionary passed to .upsert() will be unchanged. If you want to replace a record entirely, use .insert(doc, replace=True) instead.

Note that the pk and column_order parameters here are optional if you are certain that the table has already been created. You should pass them if the table may not exist at the time the first upsert is performed.

An upsert_all() method is also available, which behaves like insert_all() but performs upserts instead.


.upsert() and .upsert_all() in sqlite-utils 1.x worked like .insert(..., replace=True) and .insert_all(..., replace=True) do in 2.x. See issue #66 for details of this change.

Converting data in columns#

The table.convert(...) method can be used to apply a conversion function to the values in a column, either to update that column or to populate new columns. It is the Python library equivalent of the sqlite-utils convert command.

This feature works by registering a custom SQLite function that applies a Python transformation, then running a SQL query equivalent to UPDATE table SET column = convert_value(column);

To transform a specific column to uppercase, you would use the following:

db["dogs"].convert("name", lambda value: value.upper())

You can pass a list of columns, in which case the transformation will be applied to each one:

db["dogs"].convert(["name", "twitter"], lambda value: value.upper())

To save the output to of the transformation to a different column, use the output= parameter:

db["dogs"].convert("name", lambda value: value.upper(), output="name_upper")

This will add the new column, if it does not already exist. You can pass output_type=int or some other type to control the type of the new column - otherwise it will default to text.

If you want to drop the original column after saving the results in a separate output column, pass drop=True.

By default any rows with a falsey value for the column - such as 0 or None - will be skipped. Pass skip_false=False to disable this behaviour.

You can create multiple new columns from a single input column by passing multi=True and a conversion function that returns a Python dictionary. This example creates new upper and lower columns populated from the single title column:

    "title", lambda v: {"upper": v.upper(), "lower": v.lower()}, multi=True

The .convert() method accepts optional where= and where_args= parameters which can be used to apply the conversion to a subset of rows specified by a where clause. Here’s how to apply the conversion only to rows with an id that is higher than 20:

table.convert("title", lambda v: v.upper(), where="id > :id", where_args={"id": 20})

These behave the same as the corresponding parameters to the .rows_where() method, so you can use ? placeholders and a list of values instead of :named placeholders with a dictionary.

Working with lookup tables#

A useful pattern when populating large tables in to break common values out into lookup tables. Consider a table of Trees, where each tree has a species. Ideally these species would be split out into a separate Species table, with each one assigned an integer primary key that can be referenced from the Trees table species_id column.

Creating lookup tables explicitly#

Calling db["Species"].lookup({"name": "Palm"}) creates a table called Species (if one does not already exist) with two columns: id and name. It sets up a unique constraint on the name column to guarantee it will not contain duplicate rows. It then inserts a new row with the name set to Palm and returns the new integer primary key value.

If the Species table already exists, it will insert the new row and return the primary key. If a row with that name already exists, it will return the corresponding primary key value directly.

If you call .lookup() against an existing table without the unique constraint it will attempt to add the constraint, raising an IntegrityError if the constraint cannot be created.

If you pass in a dictionary with multiple values, both values will be used to insert or retrieve the corresponding ID and any unique constraint that is created will cover all of those columns, for example:

    "latitude": 49.1265976,
    "longitude": 2.5496218,
    "species": db["Species"].lookup({
        "common_name": "Common Juniper",
        "latin_name": "Juniperus communis"

The .lookup() method has an optional second argument which can be used to populate other columns in the table but only if the row does not exist yet. These columns will not be included in the unique index.

To create a species record with a note on when it was first seen, you can use this:

db["Species"].lookup({"name": "Palm"}, {"first_seen": "2021-03-04"})

The first time this is called the record will be created for name="Palm". Any subsequent calls with that name will ignore the second argument, even if it includes different values.

.lookup() also accepts keyword arguments, which are passed through to the insert() method and can be used to influence the shape of the created table. Supported parameters are:

  • pk - which defaults to id

  • foreign_keys

  • column_order

  • not_null

  • defaults

  • extracts

  • conversions

  • columns

  • strict

Populating lookup tables automatically during insert/upsert#

A more efficient way to work with lookup tables is to define them using the extracts= parameter, which is accepted by .insert(), .upsert(), .insert_all(), .upsert_all() and by the .table(...) factory function.

extracts= specifies columns which should be “extracted” out into a separate lookup table during the data insertion.

It can be either a list of column names, in which case the extracted table names will match the column names exactly, or it can be a dictionary mapping column names to the desired name of the extracted table.

To extract the species column out to a separate Species table, you can do this:

# Using the table factory
trees = db.table("Trees", extracts={"species": "Species"})
    "latitude": 49.1265976,
    "longitude": 2.5496218,
    "species": "Common Juniper"

# If you want the table to be called 'species', you can do this:
trees = db.table("Trees", extracts=["species"])

# Using .insert() directly
    "latitude": 49.1265976,
    "longitude": 2.5496218,
    "species": "Common Juniper"
}, extracts={"species": "Species"})

Working with many-to-many relationships#

sqlite-utils includes a shortcut for creating records using many-to-many relationships in the form of the table.m2m(...) method.

Here’s how to create two new records and connect them via a many-to-many table in a single line of code:

db["dogs"].insert({"id": 1, "name": "Cleo"}, pk="id").m2m(
    "humans", {"id": 1, "name": "Natalie"}, pk="id"

Running this example actually creates three tables: dogs, humans and a many-to-many dogs_humans table. It will insert a record into each of those tables.

The .m2m() method executes against the last record that was affected by .insert() or .update() - the record identified by the table.last_pk property. To execute .m2m() against a specific record you can first select it by passing its primary key to .update():

    "humans", {"id": 2, "name": "Simon"}, pk="id"

The first argument to .m2m() can be either the name of a table as a string or it can be the table object itself.

The second argument can be a single dictionary record or a list of dictionaries. These dictionaries will be passed to .upsert() against the specified table.

Here’s alternative code that creates the dog record and adds two people to it:

db = Database(memory=True)
dogs = db.table("dogs", pk="id")
humans = db.table("humans", pk="id")
dogs.insert({"id": 1, "name": "Cleo"}).m2m(
    humans, [
        {"id": 1, "name": "Natalie"},
        {"id": 2, "name": "Simon"}

The method will attempt to find an existing many-to-many table by looking for a table that has foreign key relationships against both of the tables in the relationship.

If it cannot find such a table, it will create a new one using the names of the two tables - dogs_humans in this example. You can customize the name of this table using the m2m_table= argument to .m2m().

It it finds multiple candidate tables with foreign keys to both of the specified tables it will raise a sqlite_utils.db.NoObviousTable exception. You can avoid this error by specifying the correct table using m2m_table=.

The .m2m() method also takes an optional pk= argument to specify the primary key that should be used if the table is created, and an optional alter=True argument to specify that any missing columns of an existing table should be added if they are needed.

Using m2m and lookup tables together#

You can work with (or create) lookup tables as part of a call to .m2m() using the lookup= parameter. This accepts the same argument as table.lookup() does - a dictionary of values that should be used to lookup or create a row in the lookup table.

This example creates a dogs table, populates it, creates a characteristics table, populates that and sets up a many-to-many relationship between the two. It chains .m2m() twice to create two associated characteristics:

db = Database(memory=True)
dogs = db.table("dogs", pk="id")
dogs.insert({"id": 1, "name": "Cleo"}).m2m(
    "characteristics", lookup={
        "name": "Playful"
    "characteristics", lookup={
        "name": "Opinionated"

You can inspect the database to see the results like this:

>>> db.table_names()
['dogs', 'characteristics', 'characteristics_dogs']
>>> list(db["dogs"].rows)
[{'id': 1, 'name': 'Cleo'}]
>>> list(db["characteristics"].rows)
[{'id': 1, 'name': 'Playful'}, {'id': 2, 'name': 'Opinionated'}]
>>> list(db["characteristics_dogs"].rows)
[{'characteristics_id': 1, 'dogs_id': 1}, {'characteristics_id': 2, 'dogs_id': 1}]
>>> print(db["characteristics_dogs"].schema)
CREATE TABLE [characteristics_dogs] (
    [characteristics_id] INTEGER REFERENCES [characteristics]([id]),
    [dogs_id] INTEGER REFERENCES [dogs]([id]),
    PRIMARY KEY ([characteristics_id], [dogs_id])

Analyzing a column#

The table.analyze_column(column) method is used by the analyze-tables CLI command.

It takes the following arguments and options:

column - required

The name of the column to analyze


The number of most common values to return. Defaults to 10.


If set to an integer, values longer than this will be truncated to this length. Defaults to None.


If set to False, the most_common field of the returned ColumnDetails will be set to None. Defaults to True.


If set to False, the least_common field of the returned ColumnDetails will be set to None. Defaults to True.

And returns a ColumnDetails named tuple with the following fields:


The name of the table


The name of the column


The total number of rows in the table


The number of rows for which this column is null


The number of rows for which this column is blank (the empty string)


The number of distinct values in this column


The N most common values as a list of (value, count) tuples`, or None if the table consists entirely of distinct values


The N least common values as a list of (value, count) tuples`, or None if the table is entirely distinct or if the number of distinct values is less than N (since they will already have been returned in most_common)

Adding columns#

You can add a new column to a table using the .add_column(col_name, col_type) method:

db["dogs"].add_column("instagram", str)
db["dogs"].add_column("weight", float)
db["dogs"].add_column("dob", datetime.date)
db["dogs"].add_column("image", "BLOB")
db["dogs"].add_column("website") # str by default

You can specify the col_type argument either using a SQLite type as a string, or by directly passing a Python type e.g. str or float.

The col_type is optional - if you omit it the type of TEXT will be used.

SQLite types you can specify are "TEXT", "INTEGER", "FLOAT" or "BLOB".

If you pass a Python type, it will be mapped to SQLite types as shown here:

float: "FLOAT"
int: "INTEGER"
bool: "INTEGER"
str: "TEXT"
bytes: "BLOB"
datetime.datetime: "TEXT"
datetime.date: "TEXT"
datetime.time: "TEXT"
datetime.timedelta: "TEXT"

# If numpy is installed
np.int8: "INTEGER"
np.int16: "INTEGER"
np.int32: "INTEGER"
np.int64: "INTEGER"
np.uint8: "INTEGER"
np.uint16: "INTEGER"
np.uint32: "INTEGER"
np.uint64: "INTEGER"
np.float16: "FLOAT"
np.float32: "FLOAT"
np.float64: "FLOAT"

You can also add a column that is a foreign key reference to another table using the fk parameter:

db["dogs"].add_column("species_id", fk="species")

This will automatically detect the name of the primary key on the species table and use that (and its type) for the new column.

You can explicitly specify the column you wish to reference using fk_col:

db["dogs"].add_column("species_id", fk="species", fk_col="ref")

You can set a NOT NULL DEFAULT 'x' constraint on the new column using not_null_default:

db["dogs"].add_column("friends_count", int, not_null_default=0)

Adding columns automatically on insert/update#

You can insert or update data that includes new columns and have the table automatically altered to fit the new schema using the alter=True argument. This can be passed to all four of .insert(), .upsert(), .insert_all() and .upsert_all(), or it can be passed to db.table(table_name, alter=True) to enable it by default for all method calls against that table instance.

db["new_table"].insert({"name": "Gareth"})
# This will throw an exception:
db["new_table"].insert({"name": "Gareth", "age": 32})
# This will succeed and add a new "age" integer column:
db["new_table"].insert({"name": "Gareth", "age": 32}, alter=True)
# You can see confirm the new column like so:
# Outputs this:
# {'name': <class 'str'>, 'age': <class 'int'>}

# This works too:
new_table = db.table("new_table", alter=True)
new_table.insert({"name": "Gareth", "age": 32, "shoe_size": 11})

Adding foreign key constraints#

The SQLite ALTER TABLE statement doesn’t have the ability to add foreign key references to an existing column.

The add_foreign_key() method here is a convenient wrapper around table.transform().

It’s also possible to add foreign keys by directly updating the sqlite_master table. The sqlite-utils-fast-fks plugin implements this pattern, using code that was included with sqlite-utils prior to version 3.35.

Here’s an example of this mechanism in action:

    {"id": 1, "name": "Sally"},
    {"id": 2, "name": "Asheesh"}
], pk="id")
    {"title": "Hedgehogs of the world", "author_id": 1},
    {"title": "How to train your wolf", "author_id": 2},
db["books"].add_foreign_key("author_id", "authors", "id")

The table.add_foreign_key(column, other_table, other_column) method takes the name of the column, the table that is being referenced and the key column within that other table. If you omit the other_column argument the primary key from that table will be used automatically. If you omit the other_table argument the table will be guessed based on some simple rules:

  • If the column is of format author_id, look for tables called author or authors

  • If the column does not end in _id, try looking for a table with the exact name of the column or that name with an added s

This method first checks that the specified foreign key references tables and columns that exist and does not clash with an existing foreign key. It will raise a sqlite_utils.db.AlterError exception if these checks fail.

To ignore the case where the key already exists, use ignore=True:

db["books"].add_foreign_key("author_id", "authors", "id", ignore=True)

Adding multiple foreign key constraints at once#

You can use db.add_foreign_keys(...) to add multiple foreign keys in one go. This method takes a list of four-tuples, each one specifying a table, column, other_table and other_column.

Here’s an example adding two foreign keys at once:

    ("dogs", "breed_id", "breeds", "id"),
    ("dogs", "home_town_id", "towns", "id")

This method runs the same checks as .add_foreign_keys() and will raise sqlite_utils.db.AlterError if those checks fail.

Adding indexes for all foreign keys#

If you want to ensure that every foreign key column in your database has a corresponding index, you can do so like this:


Dropping a table or view#

You can drop a table or view using the .drop() method:


Pass ignore=True if you want to ignore the error caused by the table or view not existing.


Transforming a table#

The SQLite ALTER TABLE statement is limited. It can add and drop columns and rename tables, but it cannot change column types, change NOT NULL status or change the primary key for a table.

The table.transform() method can do all of these things, by implementing a multi-step pattern described in the SQLite documentation:

  1. Start a transaction

  2. CREATE TABLE tablename_new_x123 with the required changes

  3. Copy the old data into the new table using INSERT INTO tablename_new_x123 SELECT * FROM tablename;

  4. DROP TABLE tablename;

  5. ALTER TABLE tablename_new_x123 RENAME TO tablename;

  6. Commit the transaction

The .transform() method takes a number of parameters, all of which are optional.

As a bonus, calling .transform() will reformat the schema for the table that is stored in SQLite to make it more readable. This works even if you call it without any arguments.

To keep the original table around instead of dropping it, pass the keep_table= option and specify the name of the table you would like it to be renamed to:

table.transform(types={"age": int}, keep_table="original_table")

Altering column types#

To alter the type of a column, use the types= argument:

# Convert the 'age' column to an integer, and 'weight' to a float
table.transform(types={"age": int, "weight": float})

See Adding columns for a list of available types.

Renaming columns#

The rename= parameter can rename columns:

# Rename 'age' to 'initial_age':
table.transform(rename={"age": "initial_age"})

Dropping columns#

To drop columns, pass them in the drop= set:

# Drop the 'age' column:

Changing primary keys#

To change the primary key for a table, use pk=. This can be passed a single column for a regular primary key, or a tuple of columns to create a compound primary key. Passing pk=None will remove the primary key and convert the table into a rowid table.

# Make `user_id` the new primary key

Changing not null status#

You can change the NOT NULL status of columns by using not_null=. You can pass this a set of columns to make those columns NOT NULL:

# Make the 'age' and 'weight' columns NOT NULL
table.transform(not_null={"age", "weight"})

If you want to take existing NOT NULL columns and change them to allow null values, you can do so by passing a dictionary of true/false values instead:

# 'age' is NOT NULL but we want to allow NULL:
table.transform(not_null={"age": False})

# Make age allow NULL and switch weight to being NOT NULL:
table.transform(not_null={"age": False, "weight": True})

Altering column defaults#

The defaults= parameter can be used to set or change the defaults for different columns:

# Set default age to 1:
table.transform(defaults={"age": 1})

# Now remove the default from that column:
table.transform(defaults={"age": None})

Changing column order#

The column_order= parameter can be used to change the order of the columns. If you pass the names of a subset of the columns those will go first and columns you omitted will appear in their existing order after them.

# Change column order
table.transform(column_order=("name", "age", "id")

Adding foreign key constraints#

You can add one or more foreign key constraints to a table using the add_foreign_keys= parameter:

        ("country", "country", "id"),
        ("continent", "continent", "id")

This accepts the same arguments described in specifying foreign keys - so you can specify them as a full tuple of (column, other_table, other_column), or you can take a shortcut and pass just the name of the column, provided the table can be automatically derived from the column name:

    add_foreign_keys=(("country", "continent"))

Replacing foreign key constraints#

The foreign_keys= parameter is similar to to add_foreign_keys= but can be be used to replace all foreign key constraints on a table, dropping any that are not explicitly mentioned:

        ("continent", "continent", "id"),

Dropping foreign key constraints#

You can use .transform() to remove foreign key constraints from a table.

This example drops two foreign keys - the one from places.country to country.id and the one from places.continent to continent.id:

    drop_foreign_keys=("country", "continent")

Custom transformations with .transform_sql()#

The .transform() method can handle most cases, but it does not automatically upgrade indexes, views or triggers associated with the table that is being transformed.

If you want to do something more advanced, you can call the table.transform_sql(...) method with the same arguments that you would have passed to table.transform(...).

This method will return a list of SQL statements that should be executed to implement the change. You can then make modifications to that SQL - or add additional SQL statements - before executing it yourself.

Extracting columns into a separate table#

The table.extract() method can be used to extract specified columns into a separate table.

Imagine a Trees table that looks like this:





52 Vine St



12 Draft St



51 Dark Ave



1252 Left St


The Species column contains duplicate values. This database could be improved by extracting that column out into a separate Species table and pointing to it using a foreign key column.

The schema of the above table is:

    [TreeAddress] TEXT,
    [Species] TEXT

Here’s how to extract the Species column using .extract():


After running this code the table schema now looks like this:

    [TreeAddress] TEXT,
    [Species_id] INTEGER,
    FOREIGN KEY(Species_id) REFERENCES Species(id)

A new Species table will have been created with the following schema:

CREATE TABLE [Species] (
    [Species] TEXT

The .extract() method defaults to creating a table with the same name as the column that was extracted, and adding a foreign key column called tablename_id.

You can specify a custom table name using table=, and a custom foreign key name using fk_column=. This example creates a table called tree_species and a foreign key column called tree_species_id:

db["Trees"].extract("Species", table="tree_species", fk_column="tree_species_id")

The resulting schema looks like this:

    [TreeAddress] TEXT,
    [tree_species_id] INTEGER,
    FOREIGN KEY(tree_species_id) REFERENCES tree_species(id)

CREATE TABLE [tree_species] (
    [Species] TEXT

You can also extract multiple columns into the same external table. Say for example you have a table like this:






52 Vine St




12 Draft St




51 Dark Ave




1252 Left St



You can pass ["CommonName", "LatinName"] to .extract() to extract both of those columns:

db["Trees"].extract(["CommonName", "LatinName"])

This produces the following schema:

    [TreeAddress] TEXT,
    [CommonName_LatinName_id] INTEGER,
    FOREIGN KEY(CommonName_LatinName_id) REFERENCES CommonName_LatinName(id)
CREATE TABLE [CommonName_LatinName] (
    [CommonName] TEXT,
    [LatinName] TEXT

The table name CommonName_LatinName is derived from the extract columns. You can use table= and fk_column= to specify custom names like this:

db["Trees"].extract(["CommonName", "LatinName"], table="Species", fk_column="species_id")

This produces the following schema:

    [TreeAddress] TEXT,
    [species_id] INTEGER,
    FOREIGN KEY(species_id) REFERENCES Species(id)
CREATE TABLE [Species] (
    [CommonName] TEXT,
    [LatinName] TEXT

You can use the rename= argument to rename columns in the lookup table. To create a Species table with columns called name and latin you can do this:

    ["CommonName", "LatinName"],
    rename={"CommonName": "name", "LatinName": "latin"}

This produces a lookup table like so:

CREATE TABLE [Species] (
    [name] TEXT,
    [latin] TEXT

Setting an ID based on the hash of the row contents#

Sometimes you will find yourself working with a dataset that includes rows that do not have a provided obvious ID, but where you would like to assign one so that you can later upsert into that table without creating duplicate records.

In these cases, a useful technique is to create an ID that is derived from the sha1 hash of the row contents.

sqlite-utils can do this for you using the hash_id= option. For example:

db = sqlite_utils.Database("dogs.db")
db["dogs"].upsert({"name": "Cleo", "twitter": "cleopaws"}, hash_id="id")


[{'id': 'f501265970505d9825d8d9f590bfab3519fb20b1', 'name': 'Cleo', 'twitter': 'cleopaws'}]

If you are going to use that ID straight away, you can access it using last_pk:

dog_id = db["dogs"].upsert({
    "name": "Cleo",
    "twitter": "cleopaws"
}, hash_id="id").last_pk
# dog_id is now "f501265970505d9825d8d9f590bfab3519fb20b1"

The hash will be created using all of the column values. To create a hash using a subset of the columns, pass the hash_id_columns= parameter:

    {"name": "Cleo", "twitter": "cleopaws", "age": 7},
    hash_id_columns=("name", "twitter")

The hash_id= parameter is optional if you specify hash_id_columns= - it will default to putting the hash in a column called id.

You can manually calculate these hashes using the hash_record(record, keys=…) utility function.

Creating views#

The .create_view() method on the database class can be used to create a view:

db.create_view("good_dogs", """
    select * from dogs where is_good_dog = 1

This will raise a sqlite_utils.utils.OperationalError if a view with that name already exists.

You can pass ignore=True to silently ignore an existing view and do nothing, or replace=True to replace an existing view with a new definition if your select statement differs from the current view:

db.create_view("good_dogs", """
    select * from dogs where is_good_dog = 1
""", replace=True)

Storing JSON#

SQLite has excellent JSON support, and sqlite-utils can help you take advantage of this: if you attempt to insert a value that can be represented as a JSON list or dictionary, sqlite-utils will create TEXT column and store your data as serialized JSON. This means you can quickly store even complex data structures in SQLite and query them using JSON features.

For example:

    "name": "The Bigfoot Discovery Museum",
    "url": "http://bigfootdiscoveryproject.com/"
    "hours": {
        "Monday": [11, 18],
        "Wednesday": [11, 18],
        "Thursday": [11, 18],
        "Friday": [11, 18],
        "Saturday": [11, 18],
        "Sunday": [11, 18]
    "address": {
        "streetAddress": "5497 Highway 9",
        "addressLocality": "Felton, CA",
        "postalCode": "95018"
    select json_extract(address, '$.addressLocality')
    from niche_museums
# Returns [('Felton, CA',)]

Converting column values using SQL functions#

Sometimes it can be useful to run values through a SQL function prior to inserting them. A simple example might be converting a value to upper case while it is being inserted.

The conversions={...} parameter can be used to specify custom SQL to be used as part of a INSERT or UPDATE SQL statement.

You can specify an upper case conversion for a specific column like so:

    "name": "The Bigfoot Discovery Museum"
}, conversions={"name": "upper(?)"})

# list(db["example"].rows) now returns:

The dictionary key is the column name to be converted. The value is the SQL fragment to use, with a ? placeholder for the original value.

A more useful example: if you are working with SpatiaLite you may find yourself wanting to create geometry values from a WKT value. Code to do that could look like this:

import sqlite3
import sqlite_utils
from shapely.geometry import shape
import httpx

db = sqlite_utils.Database("places.db")
# Initialize SpatiaLite
# Use sqlite-utils to create a places table
places = db["places"].create({"id": int, "name": str})

# Add a SpatiaLite 'geometry' column
places.add_geometry_column("geometry", "MULTIPOLYGON")

# Fetch some GeoJSON from Who's On First:
geojson = httpx.get(

# Convert to "Well Known Text" format using shapely
wkt = shape(geojson["geometry"]).wkt

# Insert the record, converting the WKT to a SpatiaLite geometry:
    {"name": "Wales", "geometry": wkt},
    conversions={"geometry": "GeomFromText(?, 4326)"},

This example uses gographical data from Who’s On First and depends on the Shapely and HTTPX Python libraries.

Checking the SQLite version#

The db.sqlite_version property returns a tuple of integers representing the version of SQLite used for that database object:

>>> db.sqlite_version
(3, 36, 0)

Dumping the database to SQL#

The db.iterdump() method returns a sequence of SQL strings representing a complete dump of the database. Use it like this:

full_sql = "".join(db.iterdump())

This uses the sqlite3.Connection.iterdump() method.

If you are using pysqlite3 or sqlean.py the underlying method may be missing. If you install the sqlite-dump package then the db.iterdump() method will use that implementation instead:

pip install sqlite-dump

Introspecting tables and views#

If you have loaded an existing table or view, you can use introspection to find out more about it:

>>> db["PlantType"]
<Table PlantType (id, value)>


The .exists() method can be used to find out if a table exists or not:

>>> db["PlantType"].exists()
>>> db["PlantType2"].exists()


The .count property shows the current number of rows (select count(*) from table):

>>> db["PlantType"].count
>>> db["Street_Tree_List"].count

This property will take advantage of Cached table counts using triggers if the use_counts_table property is set on the database. You can avoid that optimization entirely by calling table.count_where() instead of accessing the property.


The .columns property shows the columns in the table or view. It returns a list of Column(cid, name, type, notnull, default_value, is_pk) named tuples.

>>> db["PlantType"].columns
[Column(cid=0, name='id', type='INTEGER', notnull=0, default_value=None, is_pk=1),
 Column(cid=1, name='value', type='TEXT', notnull=0, default_value=None, is_pk=0)]


The .columns_dict property returns a dictionary version of the columns with just the names and Python types:

>>> db["PlantType"].columns_dict
{'id': <class 'int'>, 'value': <class 'str'>}


The .default_values property returns a dictionary of default values for each column that has a default:

>>> db["table_with_defaults"].default_values
{'score': 5}


The .pks property returns a list of strings naming the primary key columns for the table:

>>> db["PlantType"].pks

If a table has no primary keys but is a rowid table, this property will return ['rowid'].


Almost all SQLite tables have a rowid column, but a table with no explicitly defined primary keys must use that rowid as the primary key for identifying individual rows. The .use_rowid property checks to see if a table needs to use the rowid in this way - it returns True if the table has no explicitly defined primary keys and False otherwise.

>>> db["PlantType"].use_rowid


The .foreign_keys property returns any foreign key relationships for the table, as a list of ForeignKey(table, column, other_table, other_column) named tuples. It is not available on views.

>>> db["Street_Tree_List"].foreign_keys
[ForeignKey(table='Street_Tree_List', column='qLegalStatus', other_table='qLegalStatus', other_column='id'),
 ForeignKey(table='Street_Tree_List', column='qCareAssistant', other_table='qCareAssistant', other_column='id'),
 ForeignKey(table='Street_Tree_List', column='qSiteInfo', other_table='qSiteInfo', other_column='id'),
 ForeignKey(table='Street_Tree_List', column='qSpecies', other_table='qSpecies', other_column='id'),
 ForeignKey(table='Street_Tree_List', column='qCaretaker', other_table='qCaretaker', other_column='id'),
 ForeignKey(table='Street_Tree_List', column='PlantType', other_table='PlantType', other_column='id')]


The .schema property outputs the table’s schema as a SQL string:

>>> print(db["Street_Tree_List"].schema)
CREATE TABLE "Street_Tree_List" (
  "qLegalStatus" INTEGER,
  "qSpecies" INTEGER,
  "qAddress" TEXT,
  "SiteOrder" INTEGER,
  "qSiteInfo" INTEGER,
  "PlantType" INTEGER,
  "qCaretaker" INTEGER,
  "qCareAssistant" INTEGER,
  "PlantDate" TEXT,
  "PlotSize" TEXT,
  "PermitNotes" TEXT,
  "XCoord" REAL,
  "YCoord" REAL,
  "Latitude" REAL,
  "Longitude" REAL,
  "Location" TEXT
FOREIGN KEY ("PlantType") REFERENCES [PlantType](id),
    FOREIGN KEY ("qCaretaker") REFERENCES [qCaretaker](id),
    FOREIGN KEY ("qSpecies") REFERENCES [qSpecies](id),
    FOREIGN KEY ("qSiteInfo") REFERENCES [qSiteInfo](id),
    FOREIGN KEY ("qCareAssistant") REFERENCES [qCareAssistant](id),
    FOREIGN KEY ("qLegalStatus") REFERENCES [qLegalStatus](id))


The .strict property identifies if the table is a SQLite STRICT table.

>>> db["ny_times_us_counties"].strict


The .indexes property returns all indexes created for a table, as a list of Index(seq, name, unique, origin, partial, columns) named tuples. It is not available on views.

>>> db["Street_Tree_List"].indexes
[Index(seq=0, name='"Street_Tree_List_qLegalStatus"', unique=0, origin='c', partial=0, columns=['qLegalStatus']),
 Index(seq=1, name='"Street_Tree_List_qCareAssistant"', unique=0, origin='c', partial=0, columns=['qCareAssistant']),
 Index(seq=2, name='"Street_Tree_List_qSiteInfo"', unique=0, origin='c', partial=0, columns=['qSiteInfo']),
 Index(seq=3, name='"Street_Tree_List_qSpecies"', unique=0, origin='c', partial=0, columns=['qSpecies']),
 Index(seq=4, name='"Street_Tree_List_qCaretaker"', unique=0, origin='c', partial=0, columns=['qCaretaker']),
 Index(seq=5, name='"Street_Tree_List_PlantType"', unique=0, origin='c', partial=0, columns=['PlantType'])]


The .xindexes property returns more detailed information about the indexes on the table, using the SQLite PRAGMA index_xinfo() mechanism. It returns a list of XIndex(name, columns) named tuples, where columns is a list of XIndexColumn(seqno, cid, name, desc, coll, key) named tuples.

>>> db["ny_times_us_counties"].xindexes
            XIndexColumn(seqno=0, cid=0, name='date', desc=1, coll='BINARY', key=1),
            XIndexColumn(seqno=1, cid=-1, name=None, desc=0, coll='BINARY', key=0)
            XIndexColumn(seqno=0, cid=3, name='fips', desc=0, coll='BINARY', key=1),
            XIndexColumn(seqno=1, cid=-1, name=None, desc=0, coll='BINARY', key=0)


The .triggers property lists database triggers. It can be used on both database and table objects. It returns a list of Trigger(name, table, sql) named tuples.

>>> db["authors"].triggers
[Trigger(name='authors_ai', table='authors', sql='CREATE TRIGGER [authors_ai] AFTER INSERT...'),
 Trigger(name='authors_ad', table='authors', sql="CREATE TRIGGER [authors_ad] AFTER DELETE..."),
 Trigger(name='authors_au', table='authors', sql="CREATE TRIGGER [authors_au] AFTER UPDATE")]
>>> db.triggers
... similar output to db["authors"].triggers


The .triggers_dict property returns the triggers for that table as a dictionary mapping their names to their SQL definitions.

>>> db["authors"].triggers_dict
{'authors_ai': 'CREATE TRIGGER [authors_ai] AFTER INSERT...',
 'authors_ad': 'CREATE TRIGGER [authors_ad] AFTER DELETE...',
 'authors_au': 'CREATE TRIGGER [authors_au] AFTER UPDATE'}

The same property exists on the database, and will return all triggers across all tables:

>>> db.triggers_dict
{'authors_ai': 'CREATE TRIGGER [authors_ai] AFTER INSERT...',
 'authors_ad': 'CREATE TRIGGER [authors_ad] AFTER DELETE...',
 'authors_au': 'CREATE TRIGGER [authors_au] AFTER UPDATE'}


The detect_fts() method returns the associated SQLite FTS table name, if one exists for this table. If the table has not been configured for full-text search it returns None.

>>> db["authors"].detect_fts()


The .virtual_table_using property reveals if a table is a virtual table. It returns None for regular tables and the upper case version of the type of virtual table otherwise. For example:

>>> db["authors"].enable_fts(["name"])
>>> db["authors_fts"].virtual_table_using


The .has_counts_triggers property shows if a table has been configured with triggers for updating a _counts table, as described in Cached table counts using triggers.

>>> db["authors"].has_counts_triggers
>>> db["authors"].enable_counts()
>>> db["authors"].has_counts_triggers


This property on the database object returns True if the available SQLite version supports STRICT mode, which was added in SQLite 3.37.0 (on 2021-11-27).

>>> db.supports_strict

Rebuilding a full-text search table#

You can rebuild a table using the table.rebuild_fts() method. This is useful for if the table configuration changes or the indexed data has become corrupted in some way.


This method can be called on a table that has been configured for full-text search - dogs in this instance - or directly on a _fts table:


This runs the following SQL:

INSERT INTO dogs_fts (dogs_fts) VALUES ("rebuild");

Optimizing a full-text search table#

Once you have populated a FTS table you can optimize it to dramatically reduce its size like so:


This runs the following SQL:

INSERT INTO dogs_fts (dogs_fts) VALUES ("optimize");

Cached table counts using triggers#

The select count(*) query in SQLite requires a full scan of the primary key index, and can take an increasingly long time as the table grows larger.

The table.enable_counts() method can be used to configure triggers to continuously update a record in a _counts table. This value can then be used to quickly retrieve the count of rows in the associated table.


This will create the _counts table if it does not already exist, with the following schema:

CREATE TABLE [_counts] (
   [table] TEXT PRIMARY KEY,
   [count] INTEGER DEFAULT 0

You can enable cached counts for every table in a database (except for virtual tables and the _counts table itself) using the database enable_counts() method:


Once enabled, table counts will be stored in the _counts table. The count records will be automatically kept up-to-date by the triggers when rows are added or deleted to the table.

To access these counts you can query the _counts table directly or you can use the db.cached_counts() method. This method returns a dictionary mapping tables to their counts:

>>> db.cached_counts()
{'global-power-plants': 33643,
 'global-power-plants_fts_data': 136,
 'global-power-plants_fts_idx': 199,
 'global-power-plants_fts_docsize': 33643,
 'global-power-plants_fts_config': 1}

You can pass a list of table names to this method to retrieve just those counts:

>>> db.cached_counts(["global-power-plants"])
{'global-power-plants': 33643}

The table.count property executes a select count(*) query by default, unless the db.use_counts_table property is set to True.

You can set use_counts_table to True when you instantiate the database object:

db = Database("global-power-plants.db", use_counts_table=True)

If the property is True any calls to the table.count property will first attempt to find the cached count in the _counts table, and fall back on a count(*) query if the value is not available or the table is missing.

Calling the .enable_counts() method on a database or table object will set use_counts_table to True for the lifetime of that database object.

If the _counts table ever becomes out-of-sync with the actual table counts you can repair it using the .reset_counts() method:


Creating indexes#

You can create an index on a table using the .create_index(columns) method. The method takes a list of columns:


By default the index will be named idx_{table-name}_{columns}. If you pass find_unique_name=True and the automatically derived name already exists, an available name will be found by incrementing a suffix number, for example idx_items_title_2.

You can customize the name of the created index by passing the index_name parameter:

    ["is_good_dog", "age"],

To create an index in descending order for a column, wrap the column name in db.DescIndex() like this:

from sqlite_utils.db import DescIndex

    ["is_good_dog", DescIndex("age")],

You can create a unique index by passing unique=True:

db["dogs"].create_index(["name"], unique=True)

Use if_not_exists=True to do nothing if an index with that name already exists.

Pass analyze=True to run ANALYZE against the new index after creating it.

Optimizing index usage with ANALYZE#

The SQLite ANALYZE command builds a table of statistics which the query planner can use to make better decisions about which indexes to use for a given query.

You should run ANALYZE if your database is large and you do not think your indexes are being efficiently used.

To run ANALYZE against every index in a database, use this:


To run it just against a specific named index, pass the name of the index to that method:


To run against all indexes attached to a specific table, you can either pass the table name to db.analyze(...) or you can call the method directly on the table, like this:



You can optimize your database by running VACUUM against it like so:


WAL mode#

You can enable Write-Ahead Logging for a database with .enable_wal():


You can disable WAL mode using .disable_wal():


You can check the current journal mode for a database using the journal_mode property:

journal_mode = Database("my_database.db").journal_mode

This will usually be wal or delete (meaning WAL is disabled), but can have other values - see the PRAGMA journal_mode documentation.

Suggesting column types#

When you create a new table for a list of inserted or upserted Python dictionaries, those methods detect the correct types for the database columns based on the data you pass in.

In some situations you may need to intervene in this process, to customize the columns that are being created in some way - see Explicitly creating a table.

That table .create() method takes a dictionary mapping column names to the Python type they should store:

    "id": int,
    "name": str,
    "weight": float,

You can use the suggest_column_types() helper function to derive a dictionary of column names and types from a list of records, suitable to be passed to table.create().

For example:

from sqlite_utils import Database, suggest_column_types

cats = [{
    "id": 1,
    "name": "Snowflake"
}, {
    "id": 2,
    "name": "Crabtree",
    "age": 4
types = suggest_column_types(cats)
# types now looks like this:
# {"id": <class 'int'>,
#  "name": <class 'str'>,
#  "age": <class 'int'>}

# Manually add an extra field:
types["thumbnail"] = bytes
# types now looks like this:
# {"id": <class 'int'>,
#  "name": <class 'str'>,
#  "age": <class 'int'>,
#  "thumbnail": <class 'bytes'>}

# Create the table
db = Database("cats.db")
db["cats"].create(types, pk="id")
# Insert the records

# list(db["cats"].rows) now returns:
# [{"id": 1, "name": "Snowflake", "age": None, "thumbnail": None}
#  {"id": 2, "name": "Crabtree", "age": 4, "thumbnail": None}]

# The table schema looks like this:
# print(db["cats"].schema)
# CREATE TABLE [cats] (
#    [name] TEXT,
#    [age] INTEGER,
#    [thumbnail] BLOB
# )

Registering custom SQL functions#

SQLite supports registering custom SQL functions written in Python. The db.register_function() method lets you register these functions, and keeps track of functions that have already been registered.

If you use it as a method it will automatically detect the name and number of arguments needed by the function:

from sqlite_utils import Database

db = Database(memory=True)

def reverse_string(s):
    return "".join(reversed(list(s)))

print(db.execute('select reverse_string("hello")').fetchone()[0])
# This prints "olleh"

You can also use the method as a function decorator like so:

def reverse_string(s):
    return "".join(reversed(list(s)))

print(db.execute('select reverse_string("hello")').fetchone()[0])

By default, the name of the Python function will be used as the name of the SQL function. You can customize this with the name= keyword argument:

def reverse_string(s):
    return "".join(reversed(list(s)))

print(db.execute('select rev("hello")').fetchone()[0])

Python 3.8 added the ability to register deterministic SQLite functions, allowing you to indicate that a function will return the exact same result for any given inputs and hence allowing SQLite to apply some performance optimizations. You can mark a function as deterministic using deterministic=True, like this:

def reverse_string(s):
    return "".join(reversed(list(s)))

If you run this on a version of Python prior to 3.8 your code will still work, but the deterministic=True parameter will be ignored.

By default registering a function with the same name and number of arguments will have no effect - the Database instance keeps track of functions that have already been registered and skips registering them if @db.register_function is called a second time.

If you want to deliberately replace the registered function with a new implementation, use the replace=True argument:

@db.register_function(deterministic=True, replace=True)
def reverse_string(s):
    return s[::-1]

Exceptions that occur inside a user-defined function default to returning the following error:

Unexpected error: user-defined function raised exception

You can cause sqlite3 to return more useful errors, including the traceback from the custom function, by executing the following before your custom functions are executed:

from sqlite_utils.utils import sqlite3


Quoting strings for use in SQL#

In almost all cases you should pass values to your SQL queries using the optional parameters argument to db.query(), as described in Passing parameters.

If that option isn’t relevant to your use-case you can to quote a string for use with SQLite using the db.quote() method, like so:

>>> db = Database(memory=True)
>>> db.quote("hello")
>>> db.quote("hello'this'has'quotes")

Reading rows from a file#

The sqlite_utils.utils.rows_from_file() helper function can read rows (a sequence of dictionaries) from CSV, TSV, JSON or newline-delimited JSON files.

sqlite_utils.utils.rows_from_file(fp, format=None, dialect=None, encoding=None, ignore_extras=False, extras_key=None)[source]

Load a sequence of dictionaries from a file-like object containing one of four different formats.

from sqlite_utils.utils import rows_from_file
import io

rows, format = rows_from_file(io.StringIO("id,name\n1,Cleo")))
print(list(rows), format)
# Outputs [{'id': '1', 'name': 'Cleo'}] Format.CSV

This defaults to attempting to automatically detect the format of the data, or you can pass in an explicit format using the format= option.

Returns a tuple of (rows_generator, format_used) where rows_generator can be iterated over to return dictionaries, while format_used is a value from the sqlite_utils.utils.Format enum:

class Format(enum.Enum):
    CSV = 1
    TSV = 2
    JSON = 3
    NL = 4

If a CSV or TSV file includes rows with more fields than are declared in the header a sqlite_utils.utils.RowError exception will be raised when you loop over the generator.

You can instead ignore the extra data by passing ignore_extras=True.

Or pass extras_key="rest" to put those additional values in a list in a key called rest.

  • fp (BinaryIO) – a file-like object containing binary data

  • format (Format | None) – the format to use - omit this to detect the format

  • dialect (Type[Dialect] | None) – the CSV dialect to use - omit this to detect the dialect

  • encoding (str | None) – the character encoding to use when reading CSV/TSV data

  • ignore_extras (bool | None) – ignore any extra fields on rows

  • extras_key (str | None) – put any extra fields in a list with this key

Return type:

Tuple[Iterable[dict], Format]

Setting the maximum CSV field size limit#

Sometimes when working with CSV files that include extremely long fields you may see an error that looks like this:

_csv.Error: field larger than field limit (131072)

The Python standard library csv module enforces a field size limit. You can increase that limit using the csv.field_size_limit(new_limit) method (documented here) but if you don’t want to pick a new level you may instead want to increase it to the maximum possible.

The maximum possible value for this is not documented, and varies between systems.

Calling sqlite_utils.utils.maximize_csv_field_size_limit() will set the value to the highest possible for the current system:

from sqlite_utils.utils import maximize_csv_field_size_limit


If you need to reset to the original value after calling this function you can do so like this:

from sqlite_utils.utils import ORIGINAL_CSV_FIELD_SIZE_LIMIT
import csv


Detecting column types using TypeTracker#

Sometimes you may find yourself working with data that lacks type information - data from a CSV file for example.

The TypeTracker class can be used to try to automatically identify the most likely types for data that is initially represented as strings.

Consider this example:

import csv, io

csv_file = io.StringIO("id,name\n1,Cleo\n2,Cardi")
rows = list(csv.DictReader(csv_file))

# rows is now this:
# [{'id': '1', 'name': 'Cleo'}, {'id': '2', 'name': 'Cardi'}]

If we insert this data directly into a table we will get a schema that is entirely TEXT columns:

from sqlite_utils import Database

db = Database(memory=True)
# Outputs:
# CREATE TABLE [creatures] (
#    [id] TEXT,
#    [name] TEXT
# );

We can detect the best column types using a TypeTracker instance:

from sqlite_utils.utils import TypeTracker

tracker = TypeTracker()
# Outputs {'id': 'integer', 'name': 'text'}

We can then apply those types to our new table using the table.transform() method:

# Outputs:
# CREATE TABLE [creatures2] (
#    [id] INTEGER,
#    [name] TEXT
# );

SpatiaLite helpers#

SpatiaLite is a geographic extension to SQLite (similar to PostgreSQL + PostGIS). Using requires finding, loading and initializing the extension, adding geometry columns to existing tables and optionally creating spatial indexes. The utilities here help streamline that setup.

Initialize SpatiaLite#


The init_spatialite method will load and initialize the SpatiaLite extension. The path argument should be an absolute path to the compiled extension, which can be found using find_spatialite.

Returns True if SpatiaLite was successfully initialized.

from sqlite_utils.db import Database
from sqlite_utils.utils import find_spatialite

db = Database("mydb.db")

If you’ve installed SpatiaLite somewhere unexpected (for testing an alternate version, for example) you can pass in an absolute path:

from sqlite_utils.db import Database
from sqlite_utils.utils import find_spatialite

db = Database("mydb.db")

path (str | None) – Path to SpatiaLite module on disk

Return type:


Finding SpatiaLite#


The find_spatialite() function searches for the SpatiaLite SQLite extension in some common places. It returns a string path to the location, or None if SpatiaLite was not found.

You can use it in code like this:

from sqlite_utils import Database
from sqlite_utils.utils import find_spatialite

db = Database("mydb.db")
spatialite = find_spatialite()
if spatialite:

# or use with db.init_spatialite like this
Return type:

str | None

Adding geometry columns#

Table.add_geometry_column(column_name, geometry_type, srid=4326, coord_dimension='XY', not_null=False)[source]

In SpatiaLite, a geometry column can only be added to an existing table. To do so, use table.add_geometry_column, passing in a geometry type.

By default, this will add a nullable column using SRID 4326. This can be customized using the column_name, srid and not_null arguments.

Returns True if the column was successfully added, False if not.

from sqlite_utils.db import Database
from sqlite_utils.utils import find_spatialite

db = Database("mydb.db")

# the table must exist before adding a geometry column
table = db["locations"].create({"name": str})
table.add_geometry_column("geometry", "POINT")
  • column_name (str) – Name of column to add

  • geometry_type (str) – Type of geometry column, for example "GEOMETRY" or "POINT" or ``"POLYGON"

  • srid (int) – Integer SRID, defaults to 4326 for WGS84

  • coord_dimension (str) – Dimensions to use, defaults to "XY" - set to "XYZ" to work in three dimensions

  • not_null (bool) – Should the column be NOT NULL

Return type:


Creating a spatial index#


A spatial index allows for significantly faster bounding box queries. To create one, use create_spatial_index with the name of an existing geometry column.

Returns True if the index was successfully created, False if not. Calling this function if an index already exists is a no-op.

# assuming SpatiaLite is loaded, create the table, add the column
table = db["locations"].create({"name": str})
table.add_geometry_column("geometry", "POINT")

# now we can index it

# the spatial index is a virtual table, which we can inspect
# outputs:
# CREATE VIRTUAL TABLE "idx_locations_geometry" USING rtree(pkid, xmin, xmax, ymin, ymax)

column_name – Geometry column to create the spatial index against

Return type: