Skip to content

Settings Management

Pydantic Settings provides optional Pydantic features for loading a settings or config class from environment variables or secrets files.

Installation

Installation is as simple as:

pip install pydantic-settings

Usage

If you create a model that inherits from BaseSettings, the model initialiser will attempt to determine the values of any fields not passed as keyword arguments by reading from the environment. (Default values will still be used if the matching environment variable is not set.)

This makes it easy to:

  • Create a clearly-defined, type-hinted application configuration class
  • Automatically read modifications to the configuration from environment variables
  • Manually override specific settings in the initialiser where desired (e.g. in unit tests)

For example:

from typing import Any, Callable, Set

from pydantic import (
    AliasChoices,
    AmqpDsn,
    BaseModel,
    Field,
    ImportString,
    PostgresDsn,
    RedisDsn,
)

from pydantic_settings import BaseSettings, SettingsConfigDict


class SubModel(BaseModel):
    foo: str = 'bar'
    apple: int = 1


class Settings(BaseSettings):
    auth_key: str = Field(validation_alias='my_auth_key')  # (1)!

    api_key: str = Field(alias='my_api_key')  # (2)!

    redis_dsn: RedisDsn = Field(
        'redis://user:pass@localhost:6379/1',
        validation_alias=AliasChoices('service_redis_dsn', 'redis_url'),  # (3)!
    )
    pg_dsn: PostgresDsn = 'postgres://user:pass@localhost:5432/foobar'
    amqp_dsn: AmqpDsn = 'amqp://user:pass@localhost:5672/'

    special_function: ImportString[Callable[[Any], Any]] = 'math.cos'  # (4)!

    # to override domains:
    # export my_prefix_domains='["foo.com", "bar.com"]'
    domains: Set[str] = set()

    # to override more_settings:
    # export my_prefix_more_settings='{"foo": "x", "apple": 1}'
    more_settings: SubModel = SubModel()

    model_config = SettingsConfigDict(env_prefix='my_prefix_')  # (5)!


print(Settings().model_dump())
"""
{
    'auth_key': 'xxx',
    'api_key': 'xxx',
    'redis_dsn': Url('redis://user:pass@localhost:6379/1'),
    'pg_dsn': MultiHostUrl('postgres://user:pass@localhost:5432/foobar'),
    'amqp_dsn': Url('amqp://user:pass@localhost:5672/'),
    'special_function': math.cos,
    'domains': set(),
    'more_settings': {'foo': 'bar', 'apple': 1},
}
"""

  1. The environment variable name is overridden using validation_alias. In this case, the environment variable my_auth_key will be read instead of auth_key.

    Check the Field documentation for more information.

  2. The environment variable name is overridden using alias. In this case, the environment variable my_api_key will be used for both validation and serialization instead of api_key.

    Check the Field documentation for more information.

  3. The AliasChoices class allows to have multiple environment variable names for a single field. The first environment variable that is found will be used.

    Check the documentation on alias choices for more information.

  4. The ImportString class allows to import an object from a string. In this case, the environment variable special_function will be read and the function math.cos will be imported.

  5. The env_prefix config setting allows to set a prefix for all environment variables.

    Check the Environment variable names documentation for more information.

Validation of default values

Unlike pydantic BaseModel, default values of BaseSettings fields are validated by default. You can disable this behaviour by setting validate_default=False either in model_config or on field level by Field(validate_default=False):

from pydantic import Field

from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    model_config = SettingsConfigDict(validate_default=False)

    # default won't be validated
    foo: int = 'test'


print(Settings())
#> foo='test'


class Settings1(BaseSettings):
    # default won't be validated
    foo: int = Field('test', validate_default=False)


print(Settings1())
#> foo='test'

Check the Validation of default values for more information.

Environment variable names

By default, the environment variable name is the same as the field name.

You can change the prefix for all environment variables by setting the env_prefix config setting, or via the _env_prefix keyword argument on instantiation:

from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    model_config = SettingsConfigDict(env_prefix='my_prefix_')

    auth_key: str = 'xxx'  # will be read from `my_prefix_auth_key`

Note

The default env_prefix is '' (empty string).

If you want to change the environment variable name for a single field, you can use an alias.

There are two ways to do this:

  • Using Field(alias=...) (see api_key above)
  • Using Field(validation_alias=...) (see auth_key above)

Check the Field aliases documentation for more information about aliases.

env_prefix does not apply to fields with alias. It means the environment variable name is the same as field alias:

from pydantic import Field

from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    model_config = SettingsConfigDict(env_prefix='my_prefix_')

    foo: str = Field('xxx', alias='FooAlias')  # (1)!
  1. env_prefix will be ignored and the value will be read from FooAlias environment variable.

Case-sensitivity

By default, environment variable names are case-insensitive.

If you want to make environment variable names case-sensitive, you can set the case_sensitive config setting:

from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    model_config = SettingsConfigDict(case_sensitive=True)

    redis_host: str = 'localhost'

When case_sensitive is True, the environment variable names must match field names (optionally with a prefix), so in this example redis_host could only be modified via export redis_host. If you want to name environment variables all upper-case, you should name attribute all upper-case too. You can still name environment variables anything you like through Field(validation_alias=...).

Case-sensitivity can also be set via the _case_sensitive keyword argument on instantiation.

In case of nested models, the case_sensitive setting will be applied to all nested models.

import os

from pydantic import BaseModel, ValidationError

from pydantic_settings import BaseSettings


class RedisSettings(BaseModel):
    host: str
    port: int


class Settings(BaseSettings, case_sensitive=True):
    redis: RedisSettings


os.environ['redis'] = '{"host": "localhost", "port": 6379}'
print(Settings().model_dump())
#> {'redis': {'host': 'localhost', 'port': 6379}}
os.environ['redis'] = '{"HOST": "localhost", "port": 6379}'  # (1)!
try:
    Settings()
except ValidationError as e:
    print(e)
    """
    1 validation error for Settings
    redis.host
      Field required [type=missing, input_value={'HOST': 'localhost', 'port': 6379}, input_type=dict]
        For further information visit https://errors.pydantic.dev/2/v/missing
    """
  1. Note that the host field is not found because the environment variable name is HOST (all upper-case).

Note

On Windows, Python's os module always treats environment variables as case-insensitive, so the case_sensitive config setting will have no effect - settings will always be updated ignoring case.

Parsing environment variable values

By default environment variables are parsed verbatim, including if the value is empty. You can choose to ignore empty environment variables by setting the env_ignore_empty config setting to True. This can be useful if you would prefer to use the default value for a field rather than an empty value from the environment.

For most simple field types (such as int, float, str, etc.), the environment variable value is parsed the same way it would be if passed directly to the initialiser (as a string).

Complex types like list, set, dict, and sub-models are populated from the environment by treating the environment variable's value as a JSON-encoded string.

Another way to populate nested complex variables is to configure your model with the env_nested_delimiter config setting, then use an environment variable with a name pointing to the nested module fields. What it does is simply explodes your variable into nested models or dicts. So if you define a variable FOO__BAR__BAZ=123 it will convert it into FOO={'BAR': {'BAZ': 123}} If you have multiple variables with the same structure they will be merged.

Note

Sub model has to inherit from pydantic.BaseModel, Otherwise pydantic-settings will initialize sub model, collects values for sub model fields separately, and you may get unexpected results.

As an example, given the following environment variables: 

# your environment
export V0=0
export SUB_MODEL='{"v1": "json-1", "v2": "json-2"}'
export SUB_MODEL__V2=nested-2
export SUB_MODEL__V3=3
export SUB_MODEL__DEEP__V4=v4

You could load them into the following settings model:

from pydantic import BaseModel

from pydantic_settings import BaseSettings, SettingsConfigDict


class DeepSubModel(BaseModel):  # (1)!
    v4: str


class SubModel(BaseModel):  # (2)!
    v1: str
    v2: bytes
    v3: int
    deep: DeepSubModel


class Settings(BaseSettings):
    model_config = SettingsConfigDict(env_nested_delimiter='__')

    v0: str
    sub_model: SubModel


print(Settings().model_dump())
"""
{
    'v0': '0',
    'sub_model': {'v1': 'json-1', 'v2': b'nested-2', 'v3': 3, 'deep': {'v4': 'v4'}},
}
"""

  1. Sub model has to inherit from pydantic.BaseModel.

  2. Sub model has to inherit from pydantic.BaseModel.

env_nested_delimiter can be configured via the model_config as shown above, or via the _env_nested_delimiter keyword argument on instantiation.

Nested environment variables take precedence over the top-level environment variable JSON (e.g. in the example above, SUB_MODEL__V2 trumps SUB_MODEL).

You may also populate a complex type by providing your own source class.

import json
import os
from typing import Any, List, Tuple, Type

from pydantic.fields import FieldInfo

from pydantic_settings import (
    BaseSettings,
    EnvSettingsSource,
    PydanticBaseSettingsSource,
)


class MyCustomSource(EnvSettingsSource):
    def prepare_field_value(
        self, field_name: str, field: FieldInfo, value: Any, value_is_complex: bool
    ) -> Any:
        if field_name == 'numbers':
            return [int(x) for x in value.split(',')]
        return json.loads(value)


class Settings(BaseSettings):
    numbers: List[int]

    @classmethod
    def settings_customise_sources(
        cls,
        settings_cls: Type[BaseSettings],
        init_settings: PydanticBaseSettingsSource,
        env_settings: PydanticBaseSettingsSource,
        dotenv_settings: PydanticBaseSettingsSource,
        file_secret_settings: PydanticBaseSettingsSource,
    ) -> Tuple[PydanticBaseSettingsSource, ...]:
        return (MyCustomSource(settings_cls),)


os.environ['numbers'] = '1,2,3'
print(Settings().model_dump())
#> {'numbers': [1, 2, 3]}

Nested model default partial updates

By default, Pydantic settings does not allow partial updates to nested model default objects. This behavior can be overriden by setting the nested_model_default_partial_update flag to True, which will allow partial updates on nested model default object fields.

import os

from pydantic import BaseModel

from pydantic_settings import BaseSettings, SettingsConfigDict


class SubModel(BaseModel):
    val: int = 0
    flag: bool = False


class SettingsPartialUpdate(BaseSettings):
    model_config = SettingsConfigDict(
        env_nested_delimiter='__', nested_model_default_partial_update=True
    )

    nested_model: SubModel = SubModel(val=1)


class SettingsNoPartialUpdate(BaseSettings):
    model_config = SettingsConfigDict(
        env_nested_delimiter='__', nested_model_default_partial_update=False
    )

    nested_model: SubModel = SubModel(val=1)


# Apply a partial update to the default object using environment variables
os.environ['NESTED_MODEL__FLAG'] = 'True'

# When partial update is enabled, the existing SubModel instance is updated
# with nested_model.flag=True change
assert SettingsPartialUpdate().model_dump() == {
    'nested_model': {'val': 1, 'flag': True}
}

# When partial update is disabled, a new SubModel instance is instantiated
# with nested_model.flag=True change
assert SettingsNoPartialUpdate().model_dump() == {
    'nested_model': {'val': 0, 'flag': True}
}

Dotenv (.env) support

Dotenv files (generally named .env) are a common pattern that make it easy to use environment variables in a platform-independent manner.

A dotenv file follows the same general principles of all environment variables, and it looks like this:

.env
# ignore comment
ENVIRONMENT="production"
REDIS_ADDRESS=localhost:6379
MEANING_OF_LIFE=42
MY_VAR='Hello world'

Once you have your .env file filled with variables, pydantic supports loading it in two ways:

  1. Setting the env_file (and env_file_encoding if you don't want the default encoding of your OS) on model_config in the BaseSettings class: 
    from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    model_config = SettingsConfigDict(env_file='.env', env_file_encoding='utf-8')
  2. Instantiating the BaseSettings derived class with the _env_file keyword argument (and the _env_file_encoding if needed): 
    from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    model_config = SettingsConfigDict(env_file='.env', env_file_encoding='utf-8')


settings = Settings(_env_file='prod.env', _env_file_encoding='utf-8')
    In either case, the value of the passed argument can be any valid path or filename, either absolute or relative to the current working directory. From there, pydantic will handle everything for you by loading in your variables and validating them.

Note

If a filename is specified for env_file, Pydantic will only check the current working directory and won't check any parent directories for the .env file.

Even when using a dotenv file, pydantic will still read environment variables as well as the dotenv file, environment variables will always take priority over values loaded from a dotenv file.

Passing a file path via the _env_file keyword argument on instantiation (method 2) will override the value (if any) set on the model_config class. If the above snippets were used in conjunction, prod.env would be loaded while .env would be ignored.

If you need to load multiple dotenv files, you can pass multiple file paths as a tuple or list. The files will be loaded in order, with each file overriding the previous one.

from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    model_config = SettingsConfigDict(
        # `.env.prod` takes priority over `.env`
        env_file=('.env', '.env.prod')
    )

You can also use the keyword argument override to tell Pydantic not to load any file at all (even if one is set in the model_config class) by passing None as the instantiation keyword argument, e.g. settings = Settings(_env_file=None).

Because python-dotenv is used to parse the file, bash-like semantics such as export can be used which (depending on your OS and environment) may allow your dotenv file to also be used with source, see python-dotenv's documentation for more details.

Pydantic settings consider extra config in case of dotenv file. It means if you set the extra=forbid (default) on model_config and your dotenv file contains an entry for a field that is not defined in settings model, it will raise ValidationError in settings construction.

For compatibility with pydantic 1.x BaseSettings you should use extra=ignore: 

from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    model_config = SettingsConfigDict(env_file='.env', extra='ignore')

Note

Pydantic settings loads all the values from dotenv file and passes it to the model, regardless of the model's env_prefix. So if you provide extra values in a dotenv file, whether they start with env_prefix or not, a ValidationError will be raised.

Command Line Support

Pydantic settings provides integrated CLI support, making it easy to quickly define CLI applications using Pydantic models. There are two primary use cases for Pydantic settings CLI:

  1. When using a CLI to override fields in Pydantic models.
  2. When using Pydantic models to define CLIs.

By default, the experience is tailored towards use case #1 and builds on the foundations established in parsing environment variables. If your use case primarily falls into #2, you will likely want to enable most of the defaults outlined at the end of creating CLI applications.

The Basics

To get started, let's revisit the example presented in parsing environment variables but using a Pydantic settings CLI:

import sys

from pydantic import BaseModel

from pydantic_settings import BaseSettings, SettingsConfigDict


class DeepSubModel(BaseModel):
    v4: str


class SubModel(BaseModel):
    v1: str
    v2: bytes
    v3: int
    deep: DeepSubModel


class Settings(BaseSettings):
    model_config = SettingsConfigDict(cli_parse_args=True)

    v0: str
    sub_model: SubModel


sys.argv = [
    'example.py',
    '--v0=0',
    '--sub_model={"v1": "json-1", "v2": "json-2"}',
    '--sub_model.v2=nested-2',
    '--sub_model.v3=3',
    '--sub_model.deep.v4=v4',
]

print(Settings().model_dump())
"""
{
    'v0': '0',
    'sub_model': {'v1': 'json-1', 'v2': b'nested-2', 'v3': 3, 'deep': {'v4': 'v4'}},
}
"""

To enable CLI parsing, we simply set the cli_parse_args flag to a valid value, which retains similar conotations as defined in argparse.

Note that a CLI settings source is the topmost source by default unless its priority value is customised:

import os
import sys
from typing import Tuple, Type

from pydantic_settings import (
    BaseSettings,
    CliSettingsSource,
    PydanticBaseSettingsSource,
)


class Settings(BaseSettings):
    my_foo: str

    @classmethod
    def settings_customise_sources(
        cls,
        settings_cls: Type[BaseSettings],
        init_settings: PydanticBaseSettingsSource,
        env_settings: PydanticBaseSettingsSource,
        dotenv_settings: PydanticBaseSettingsSource,
        file_secret_settings: PydanticBaseSettingsSource,
    ) -> Tuple[PydanticBaseSettingsSource, ...]:
        return env_settings, CliSettingsSource(settings_cls, cli_parse_args=True)


os.environ['MY_FOO'] = 'from environment'

sys.argv = ['example.py', '--my_foo=from cli']

print(Settings().model_dump())
#> {'my_foo': 'from environment'}

Lists

CLI argument parsing of lists supports intermixing of any of the below three styles:

  • JSON style --field='[1,2]'
  • Argparse style --field 1 --field 2
  • Lazy style --field=1,2
import sys
from typing import List

from pydantic_settings import BaseSettings


class Settings(BaseSettings, cli_parse_args=True):
    my_list: List[int]


sys.argv = ['example.py', '--my_list', '[1,2]']
print(Settings().model_dump())
#> {'my_list': [1, 2]}

sys.argv = ['example.py', '--my_list', '1', '--my_list', '2']
print(Settings().model_dump())
#> {'my_list': [1, 2]}

sys.argv = ['example.py', '--my_list', '1,2']
print(Settings().model_dump())
#> {'my_list': [1, 2]}

Dictionaries

CLI argument parsing of dictionaries supports intermixing of any of the below two styles:

  • JSON style --field='{"k1": 1, "k2": 2}'
  • Environment variable style --field k1=1 --field k2=2

These can be used in conjunction with list forms as well, e.g:

  • --field k1=1,k2=2 --field k3=3 --field '{"k4": 4}' etc.
import sys
from typing import Dict

from pydantic_settings import BaseSettings


class Settings(BaseSettings, cli_parse_args=True):
    my_dict: Dict[str, int]


sys.argv = ['example.py', '--my_dict', '{"k1":1,"k2":2}']
print(Settings().model_dump())
#> {'my_dict': {'k1': 1, 'k2': 2}}

sys.argv = ['example.py', '--my_dict', 'k1=1', '--my_dict', 'k2=2']
print(Settings().model_dump())
#> {'my_dict': {'k1': 1, 'k2': 2}}

Literals and Enums

CLI argument parsing of literals and enums are converted into CLI choices.

import sys
from enum import IntEnum
from typing import Literal

from pydantic_settings import BaseSettings


class Fruit(IntEnum):
    pear = 0
    kiwi = 1
    lime = 2


class Settings(BaseSettings, cli_parse_args=True):
    fruit: Fruit
    pet: Literal['dog', 'cat', 'bird']


sys.argv = ['example.py', '--fruit', 'lime', '--pet', 'cat']
print(Settings().model_dump())
#> {'fruit': <Fruit.lime: 2>, 'pet': 'cat'}

Aliases

Pydantic field aliases are added as CLI argument aliases. Aliases of length one are converted into short options.

import sys

from pydantic import AliasChoices, AliasPath, Field

from pydantic_settings import BaseSettings


class User(BaseSettings, cli_parse_args=True):
    first_name: str = Field(
        validation_alias=AliasChoices('f', 'fname', AliasPath('name', 0))
    )
    last_name: str = Field(
        validation_alias=AliasChoices('l', 'lname', AliasPath('name', 1))
    )


sys.argv = ['example.py', '--fname', 'John', '--lname', 'Doe']
print(User().model_dump())
#> {'first_name': 'John', 'last_name': 'Doe'}

sys.argv = ['example.py', '-f', 'John', '-l', 'Doe']
print(User().model_dump())
#> {'first_name': 'John', 'last_name': 'Doe'}

sys.argv = ['example.py', '--name', 'John,Doe']
print(User().model_dump())
#> {'first_name': 'John', 'last_name': 'Doe'}

sys.argv = ['example.py', '--name', 'John', '--lname', 'Doe']
print(User().model_dump())
#> {'first_name': 'John', 'last_name': 'Doe'}

Subcommands and Positional Arguments

Subcommands and positional arguments are expressed using the CliSubCommand and CliPositionalArg annotations. These annotations can only be applied to required fields (i.e. fields that do not have a default value). Furthermore, subcommands must be a valid type derived from either a pydantic BaseModel or pydantic.dataclasses dataclass.

Parsed subcommands can be retrieved from model instances using the get_subcommand utility function. If a subcommand is not required, set the is_required flag to False to disable raising an error if no subcommand is found.

Note

CLI settings subcommands are limited to a single subparser per model. In other words, all subcommands for a model are grouped under a single subparser; it does not allow for multiple subparsers with each subparser having its own set of subcommands. For more information on subparsers, see argparse subcommands.

Note

CliSubCommand and CliPositionalArg are always case sensitive.

import sys

from pydantic import BaseModel

from pydantic_settings import (
    BaseSettings,
    CliPositionalArg,
    CliSubCommand,
    SettingsError,
    get_subcommand,
)


class Init(BaseModel):
    directory: CliPositionalArg[str]


class Clone(BaseModel):
    repository: CliPositionalArg[str]
    directory: CliPositionalArg[str]


class Git(BaseSettings, cli_parse_args=True, cli_exit_on_error=False):
    clone: CliSubCommand[Clone]
    init: CliSubCommand[Init]


# Run without subcommands
sys.argv = ['example.py']
cmd = Git()
assert cmd.model_dump() == {'clone': None, 'init': None}

try:
    # Will raise an error since no subcommand was provided
    get_subcommand(cmd).model_dump()
except SettingsError as err:
    assert str(err) == 'Error: CLI subcommand is required {clone, init}'

# Will not raise an error since subcommand is not required
assert get_subcommand(cmd, is_required=False) is None


# Run the clone subcommand
sys.argv = ['example.py', 'clone', 'repo', 'dest']
cmd = Git()
assert cmd.model_dump() == {
    'clone': {'repository': 'repo', 'directory': 'dest'},
    'init': None,
}

# Returns the subcommand model instance (in this case, 'clone')
assert get_subcommand(cmd).model_dump() == {
    'directory': 'dest',
    'repository': 'repo',
}

The CliSubCommand and CliPositionalArg annotations also support union operations and aliases. For unions of Pydantic models, it is important to remember the nuances that can arise during validation. Specifically, for unions of subcommands that are identical in content, it is recommended to break them out into separate CliSubCommand fields to avoid any complications. Lastly, the derived subcommand names from unions will be the names of the Pydantic model classes themselves.

When assigning aliases to CliSubCommand or CliPositionalArg fields, only a single alias can be assigned. For non-union subcommands, aliasing will change the displayed help text and subcommand name. Conversely, for union subcommands, aliasing will have no tangible effect from the perspective of the CLI settings source. Lastly, for positional arguments, aliasing will change the CLI help text displayed for the field.

import sys
from typing import Union

from pydantic import BaseModel, Field

from pydantic_settings import (
    BaseSettings,
    CliPositionalArg,
    CliSubCommand,
    get_subcommand,
)


class Alpha(BaseModel):
    """Apha Help"""

    cmd_alpha: CliPositionalArg[str] = Field(alias='alpha-cmd')


class Beta(BaseModel):
    """Beta Help"""

    opt_beta: str = Field(alias='opt-beta')


class Gamma(BaseModel):
    """Gamma Help"""

    opt_gamma: str = Field(alias='opt-gamma')


class Root(BaseSettings, cli_parse_args=True, cli_exit_on_error=False):
    alpha_or_beta: CliSubCommand[Union[Alpha, Beta]] = Field(alias='alpha-or-beta-cmd')
    gamma: CliSubCommand[Gamma] = Field(alias='gamma-cmd')


sys.argv = ['example.py', 'Alpha', 'hello']
assert get_subcommand(Root()).model_dump() == {'cmd_alpha': 'hello'}

sys.argv = ['example.py', 'Beta', '--opt-beta=hey']
assert get_subcommand(Root()).model_dump() == {'opt_beta': 'hey'}

sys.argv = ['example.py', 'gamma-cmd', '--opt-gamma=hi']
assert get_subcommand(Root()).model_dump() == {'opt_gamma': 'hi'}

Creating CLI Applications

The CliApp class provides two utility methods, CliApp.run and CliApp.run_subcommand, that can be used to run a Pydantic BaseSettings, BaseModel, or pydantic.dataclasses.dataclass as a CLI application. Primarily, the methods provide structure for running cli_cmd methods associated with models.

CliApp.run can be used in directly providing the cli_args to be parsed, and will run the model cli_cmd method (if defined) after instantiation:

from pydantic_settings import BaseSettings, CliApp


class Settings(BaseSettings):
    this_foo: str

    def cli_cmd(self) -> None:
        # Print the parsed data
        print(self.model_dump())
        #> {'this_foo': 'is such a foo'}

        # Update the parsed data showing cli_cmd ran
        self.this_foo = 'ran the foo cli cmd'


s = CliApp.run(Settings, cli_args=['--this_foo', 'is such a foo'])
print(s.model_dump())
#> {'this_foo': 'ran the foo cli cmd'}

Similarly, the CliApp.run_subcommand can be used in recursive fashion to run the cli_cmd method of a subcommand:

from pydantic import BaseModel

from pydantic_settings import CliApp, CliPositionalArg, CliSubCommand


class Init(BaseModel):
    directory: CliPositionalArg[str]

    def cli_cmd(self) -> None:
        print(f'git init "{self.directory}"')
        #> git init "dir"
        self.directory = 'ran the git init cli cmd'


class Clone(BaseModel):
    repository: CliPositionalArg[str]
    directory: CliPositionalArg[str]

    def cli_cmd(self) -> None:
        print(f'git clone from "{self.repository}" into "{self.directory}"')
        self.directory = 'ran the clone cli cmd'


class Git(BaseModel):
    clone: CliSubCommand[Clone]
    init: CliSubCommand[Init]

    def cli_cmd(self) -> None:
        CliApp.run_subcommand(self)


cmd = CliApp.run(Git, cli_args=['init', 'dir'])
assert cmd.model_dump() == {
    'clone': None,
    'init': {'directory': 'ran the git init cli cmd'},
}

Note

Unlike CliApp.run, CliApp.run_subcommand requires the subcommand model to have a defined cli_cmd method.

For BaseModel and pydantic.dataclasses.dataclass types, CliApp.run will internally use the following BaseSettings configuration defaults:

  • alias_generator=AliasGenerator(lambda s: s.replace('_', '-'))
  • nested_model_default_partial_update=True
  • case_sensitive=True
  • cli_hide_none_type=True
  • cli_avoid_json=True
  • cli_enforce_required=True
  • cli_implicit_flags=True

Note

The alias generator for kebab case does not propagate to subcommands or submodels and will have to be manually set in these cases.

Mutually Exclusive Groups

CLI mutually exclusive groups can be created by inheriting from the CliMutuallyExclusiveGroup class.

Note

A CliMutuallyExclusiveGroup cannot be used in a union or contain nested models.

from typing import Optional

from pydantic import BaseModel

from pydantic_settings import CliApp, CliMutuallyExclusiveGroup, SettingsError


class Circle(CliMutuallyExclusiveGroup):
    radius: Optional[float] = None
    diameter: Optional[float] = None
    perimeter: Optional[float] = None


class Settings(BaseModel):
    circle: Circle


try:
    CliApp.run(
        Settings,
        cli_args=['--circle.radius=1', '--circle.diameter=2'],
        cli_exit_on_error=False,
    )
except SettingsError as e:
    print(e)
    """
    error parsing CLI: argument --circle.diameter: not allowed with argument --circle.radius
    """

Customizing the CLI Experience

The below flags can be used to customise the CLI experience to your needs.

Change the Displayed Program Name

Change the default program name displayed in the help text usage by setting cli_prog_name. By default, it will derive the name of the currently executing program from sys.argv[0], just like argparse.

import sys

from pydantic_settings import BaseSettings


class Settings(BaseSettings, cli_parse_args=True, cli_prog_name='appdantic'):
    pass


try:
    sys.argv = ['example.py', '--help']
    Settings()
except SystemExit as e:
    print(e)
    #> 0
"""
usage: appdantic [-h]

options:
  -h, --help  show this help message and exit
"""

CLI Boolean Flags

Change whether boolean fields should be explicit or implicit by default using the cli_implicit_flags setting. By default, boolean fields are "explicit", meaning a boolean value must be explicitly provided on the CLI, e.g. --flag=True. Conversely, boolean fields that are "implicit" derive the value from the flag itself, e.g. --flag,--no-flag, which removes the need for an explicit value to be passed.

Additionally, the provided CliImplicitFlag and CliExplicitFlag annotations can be used for more granular control when necessary.

Note

For python < 3.9 the --no-flag option is not generated due to an underlying argparse limitation.

Note

For python < 3.9 the CliImplicitFlag and CliExplicitFlag annotations can only be applied to optional boolean fields.

from pydantic_settings import BaseSettings, CliExplicitFlag, CliImplicitFlag


class ExplicitSettings(BaseSettings, cli_parse_args=True):
    """Boolean fields are explicit by default."""

    explicit_req: bool
    """
    --explicit_req bool   (required)
    """

    explicit_opt: bool = False
    """
    --explicit_opt bool   (default: False)
    """

    # Booleans are explicit by default, so must override implicit flags with annotation
    implicit_req: CliImplicitFlag[bool]
    """
    --implicit_req, --no-implicit_req (required)
    """

    implicit_opt: CliImplicitFlag[bool] = False
    """
    --implicit_opt, --no-implicit_opt (default: False)
    """


class ImplicitSettings(BaseSettings, cli_parse_args=True, cli_implicit_flags=True):
    """With cli_implicit_flags=True, boolean fields are implicit by default."""

    # Booleans are implicit by default, so must override explicit flags with annotation
    explicit_req: CliExplicitFlag[bool]
    """
    --explicit_req bool   (required)
    """

    explicit_opt: CliExplicitFlag[bool] = False
    """
    --explicit_opt bool   (default: False)
    """

    implicit_req: bool
    """
    --implicit_req, --no-implicit_req (required)
    """

    implicit_opt: bool = False
    """
    --implicit_opt, --no-implicit_opt (default: False)
    """

Ignore Unknown Arguments

Change whether to ignore unknown CLI arguments and only parse known ones using cli_ignore_unknown_args. By default, the CLI does not ignore any args.

import sys

from pydantic_settings import BaseSettings


class Settings(BaseSettings, cli_parse_args=True, cli_ignore_unknown_args=True):
    good_arg: str


sys.argv = ['example.py', '--bad-arg=bad', 'ANOTHER_BAD_ARG', '--good_arg=hello world']
print(Settings().model_dump())
#> {'good_arg': 'hello world'}

Change Whether CLI Should Exit on Error

Change whether the CLI internal parser will exit on error or raise a SettingsError exception by using cli_exit_on_error. By default, the CLI internal parser will exit on error.

import sys

from pydantic_settings import BaseSettings, SettingsError


class Settings(BaseSettings, cli_parse_args=True, cli_exit_on_error=False): ...


try:
    sys.argv = ['example.py', '--bad-arg']
    Settings()
except SettingsError as e:
    print(e)
    #> error parsing CLI: unrecognized arguments: --bad-arg

Enforce Required Arguments at CLI

Pydantic settings is designed to pull values in from various sources when instantating a model. This means a field that is required is not strictly required from any single source (e.g. the CLI). Instead, all that matters is that one of the sources provides the required value.

However, if your use case aligns more with #2, using Pydantic models to define CLIs, you will likely want required fields to be strictly required at the CLI. We can enable this behavior by using cli_enforce_required.

import os
import sys

from pydantic import Field

from pydantic_settings import BaseSettings, SettingsError


class Settings(
    BaseSettings,
    cli_parse_args=True,
    cli_enforce_required=True,
    cli_exit_on_error=False,
):
    my_required_field: str = Field(description='a top level required field')


os.environ['MY_REQUIRED_FIELD'] = 'hello from environment'

try:
    sys.argv = ['example.py']
    Settings()
except SettingsError as e:
    print(e)
    #> error parsing CLI: the following arguments are required: --my_required_field

Change the None Type Parse String

Change the CLI string value that will be parsed (e.g. "null", "void", "None", etc.) into None by setting cli_parse_none_str. By default it will use the env_parse_none_str value if set. Otherwise, it will default to "null" if cli_avoid_json is False, and "None" if cli_avoid_json is True.

import sys
from typing import Optional

from pydantic import Field

from pydantic_settings import BaseSettings


class Settings(BaseSettings, cli_parse_args=True, cli_parse_none_str='void'):
    v1: Optional[int] = Field(description='the top level v0 option')


sys.argv = ['example.py', '--v1', 'void']
print(Settings().model_dump())
#> {'v1': None}

Hide None Type Values

Hide None values from the CLI help text by enabling cli_hide_none_type.

import sys
from typing import Optional

from pydantic import Field

from pydantic_settings import BaseSettings


class Settings(BaseSettings, cli_parse_args=True, cli_hide_none_type=True):
    v0: Optional[str] = Field(description='the top level v0 option')


try:
    sys.argv = ['example.py', '--help']
    Settings()
except SystemExit as e:
    print(e)
    #> 0
"""
usage: example.py [-h] [--v0 str]

options:
  -h, --help  show this help message and exit
  --v0 str    the top level v0 option (required)
"""

Avoid Adding JSON CLI Options

Avoid adding complex fields that result in JSON strings at the CLI by enabling cli_avoid_json.

import sys

from pydantic import BaseModel, Field

from pydantic_settings import BaseSettings


class SubModel(BaseModel):
    v1: int = Field(description='the sub model v1 option')


class Settings(BaseSettings, cli_parse_args=True, cli_avoid_json=True):
    sub_model: SubModel = Field(
        description='The help summary for SubModel related options'
    )


try:
    sys.argv = ['example.py', '--help']
    Settings()
except SystemExit as e:
    print(e)
    #> 0
"""
usage: example.py [-h] [--sub_model.v1 int]

options:
  -h, --help          show this help message and exit

sub_model options:
  The help summary for SubModel related options

  --sub_model.v1 int  the sub model v1 option (required)
"""

Use Class Docstring for Group Help Text

By default, when populating the group help text for nested models it will pull from the field descriptions. Alternatively, we can also configure CLI settings to pull from the class docstring instead.

Note

If the field is a union of nested models the group help text will always be pulled from the field description; even if cli_use_class_docs_for_groups is set to True.

import sys

from pydantic import BaseModel, Field

from pydantic_settings import BaseSettings


class SubModel(BaseModel):
    """The help text from the class docstring."""

    v1: int = Field(description='the sub model v1 option')


class Settings(BaseSettings, cli_parse_args=True, cli_use_class_docs_for_groups=True):
    """My application help text."""

    sub_model: SubModel = Field(description='The help text from the field description')


try:
    sys.argv = ['example.py', '--help']
    Settings()
except SystemExit as e:
    print(e)
    #> 0
"""
usage: example.py [-h] [--sub_model JSON] [--sub_model.v1 int]

My application help text.

options:
  -h, --help          show this help message and exit

sub_model options:
  The help text from the class docstring.

  --sub_model JSON    set sub_model from JSON string
  --sub_model.v1 int  the sub model v1 option (required)
"""

Change the CLI Flag Prefix Character

Change The CLI flag prefix character used in CLI optional arguments by settings cli_flag_prefix_char.

import sys

from pydantic import AliasChoices, Field

from pydantic_settings import BaseSettings


class Settings(BaseSettings, cli_parse_args=True, cli_flag_prefix_char='+'):
    my_arg: str = Field(validation_alias=AliasChoices('m', 'my-arg'))


sys.argv = ['example.py', '++my-arg', 'hi']
print(Settings().model_dump())
#> {'my_arg': 'hi'}

sys.argv = ['example.py', '+m', 'hi']
print(Settings().model_dump())
#> {'my_arg': 'hi'}

Suppressing Fields from CLI Help Text

To suppress a field from the CLI help text, the CliSuppress annotation can be used for field types, or the CLI_SUPPRESS string constant can be used for field descriptions.

import sys

from pydantic import Field

from pydantic_settings import CLI_SUPPRESS, BaseSettings, CliSuppress


class Settings(BaseSettings, cli_parse_args=True):
    """Suppress fields from CLI help text."""

    field_a: CliSuppress[int] = 0
    field_b: str = Field(default=1, description=CLI_SUPPRESS)


try:
    sys.argv = ['example.py', '--help']
    Settings()
except SystemExit as e:
    print(e)
    #> 0
"""
usage: example.py [-h]

Suppress fields from CLI help text.

options:
  -h, --help          show this help message and exit
"""

Integrating with Existing Parsers

A CLI settings source can be integrated with existing parsers by overriding the default CLI settings source with a user defined one that specifies the root_parser object.

import sys
from argparse import ArgumentParser

from pydantic_settings import BaseSettings, CliApp, CliSettingsSource

parser = ArgumentParser()
parser.add_argument('--food', choices=['pear', 'kiwi', 'lime'])


class Settings(BaseSettings):
    name: str = 'Bob'


# Set existing `parser` as the `root_parser` object for the user defined settings source
cli_settings = CliSettingsSource(Settings, root_parser=parser)

# Parse and load CLI settings from the command line into the settings source.
sys.argv = ['example.py', '--food', 'kiwi', '--name', 'waldo']
s = CliApp.run(Settings, cli_settings_source=cli_settings)
print(s.model_dump())
#> {'name': 'waldo'}

# Load CLI settings from pre-parsed arguments. i.e., the parsing occurs elsewhere and we
# just need to load the pre-parsed args into the settings source.
parsed_args = parser.parse_args(['--food', 'kiwi', '--name', 'ralph'])
s = CliApp.run(Settings, cli_args=parsed_args, cli_settings_source=cli_settings)
print(s.model_dump())
#> {'name': 'ralph'}

A CliSettingsSource connects with a root_parser object by using parser methods to add settings_cls fields as command line arguments. The CliSettingsSource internal parser representation is based on the argparse library, and therefore, requires parser methods that support the same attributes as their argparse counterparts. The available parser methods that can be customised, along with their argparse counterparts (the defaults), are listed below:

  • parse_args_method - (argparse.ArgumentParser.parse_args)
  • add_argument_method - (argparse.ArgumentParser.add_argument)
  • add_argument_group_method - (argparse.ArgumentParser.add_argument_group)
  • add_parser_method - (argparse._SubParsersAction.add_parser)
  • add_subparsers_method - (argparse.ArgumentParser.add_subparsers)
  • formatter_class - (argparse.RawDescriptionHelpFormatter)

For a non-argparse parser the parser methods can be set to None if not supported. The CLI settings will only raise an error when connecting to the root parser if a parser method is necessary but set to None.

Note

The formatter_class is only applied to subcommands. The CliSettingsSource never touches or modifies any of the external parser settings to avoid breaking changes. Since subcommands reside on their own internal parser trees, we can safely apply the formatter_class settings without breaking the external parser logic.

Secrets

Placing secret values in files is a common pattern to provide sensitive configuration to an application.

A secret file follows the same principal as a dotenv file except it only contains a single value and the file name is used as the key. A secret file will look like the following:

/var/run/database_password
super_secret_database_password

Once you have your secret files, pydantic supports loading it in two ways:

  1. Setting the secrets_dir on model_config in a BaseSettings class to the directory where your secret files are stored. 
    from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    model_config = SettingsConfigDict(secrets_dir='/var/run')

    database_password: str
  2. Instantiating the BaseSettings derived class with the _secrets_dir keyword argument: 
    settings = Settings(_secrets_dir='/var/run')

In either case, the value of the passed argument can be any valid directory, either absolute or relative to the current working directory. Note that a non existent directory will only generate a warning. From there, pydantic will handle everything for you by loading in your variables and validating them.

Even when using a secrets directory, pydantic will still read environment variables from a dotenv file or the environment, a dotenv file and environment variables will always take priority over values loaded from the secrets directory.

Passing a file path via the _secrets_dir keyword argument on instantiation (method 2) will override the value (if any) set on the model_config class.

If you need to load settings from multiple secrets directories, you can pass multiple paths as a tuple or list. Just like for env_file, values from subsequent paths override previous ones.

from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    # files in '/run/secrets' take priority over '/var/run'
    model_config = SettingsConfigDict(secrets_dir=('/var/run', '/run/secrets'))

    database_password: str

If any of secrets_dir is missing, it is ignored, and warning is shown. If any of secrets_dir is a file, error is raised.

Use Case: Docker Secrets

Docker Secrets can be used to provide sensitive configuration to an application running in a Docker container. To use these secrets in a pydantic application the process is simple. More information regarding creating, managing and using secrets in Docker see the official Docker documentation.

First, define your Settings class with a SettingsConfigDict that specifies the secrets directory.

from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    model_config = SettingsConfigDict(secrets_dir='/run/secrets')

    my_secret_data: str

Note

By default Docker uses /run/secrets as the target mount point. If you want to use a different location, change Config.secrets_dir accordingly.

Then, create your secret via the Docker CLI 

printf "This is a secret" | docker secret create my_secret_data -

Last, run your application inside a Docker container and supply your newly created secret 

docker service create --name pydantic-with-secrets --secret my_secret_data pydantic-app:latest

Azure Key Vault

You must set two parameters:

  • url: For example, https://my-resource.vault.azure.net/.
  • credential: If you use DefaultAzureCredential, in local you can execute az login to get your identity credentials. The identity must have a role assignment (the recommended one is Key Vault Secrets User), so you can access the secrets.

You must have the same naming convention in the field name as in the Key Vault secret name. For example, if the secret is named SqlServerPassword, the field name must be the same. You can use an alias too.

In Key Vault, nested models are supported with the -- separator. For example, SqlServer--Password.

Key Vault arrays (e.g. MySecret--0, MySecret--1) are not supported.

import os
from typing import Tuple, Type

from azure.identity import DefaultAzureCredential
from pydantic import BaseModel

from pydantic_settings import (
    AzureKeyVaultSettingsSource,
    BaseSettings,
    PydanticBaseSettingsSource,
)


class SubModel(BaseModel):
    a: str


class AzureKeyVaultSettings(BaseSettings):
    foo: str
    bar: int
    sub: SubModel

    @classmethod
    def settings_customise_sources(
        cls,
        settings_cls: Type[BaseSettings],
        init_settings: PydanticBaseSettingsSource,
        env_settings: PydanticBaseSettingsSource,
        dotenv_settings: PydanticBaseSettingsSource,
        file_secret_settings: PydanticBaseSettingsSource,
    ) -> Tuple[PydanticBaseSettingsSource, ...]:
        az_key_vault_settings = AzureKeyVaultSettingsSource(
            settings_cls,
            os.environ['AZURE_KEY_VAULT_URL'],
            DefaultAzureCredential(),
        )
        return (
            init_settings,
            env_settings,
            dotenv_settings,
            file_secret_settings,
            az_key_vault_settings,
        )

Other settings source

Other settings sources are available for common configuration files:

  • JsonConfigSettingsSource using json_file and json_file_encoding arguments
  • PyprojectTomlConfigSettingsSource using (optional) pyproject_toml_depth and (optional) pyproject_toml_table_header arguments
  • TomlConfigSettingsSource using toml_file argument
  • YamlConfigSettingsSource using yaml_file and yaml_file_encoding arguments

You can also provide multiple files by providing a list of path: 

toml_file = ['config.default.toml', 'config.custom.toml']
To use them, you can use the same mechanism described here

from typing import Tuple, Type

from pydantic import BaseModel

from pydantic_settings import (
    BaseSettings,
    PydanticBaseSettingsSource,
    SettingsConfigDict,
    TomlConfigSettingsSource,
)


class Nested(BaseModel):
    nested_field: str


class Settings(BaseSettings):
    foobar: str
    nested: Nested
    model_config = SettingsConfigDict(toml_file='config.toml')

    @classmethod
    def settings_customise_sources(
        cls,
        settings_cls: Type[BaseSettings],
        init_settings: PydanticBaseSettingsSource,
        env_settings: PydanticBaseSettingsSource,
        dotenv_settings: PydanticBaseSettingsSource,
        file_secret_settings: PydanticBaseSettingsSource,
    ) -> Tuple[PydanticBaseSettingsSource, ...]:
        return (TomlConfigSettingsSource(settings_cls),)

This will be able to read the following "config.toml" file, located in your working directory:

foobar = "Hello"
[nested]
nested_field = "world!"

pyproject.toml

"pyproject.toml" is a standardized file for providing configuration values in Python projects. PEP 518 defines a [tool] table that can be used to provide arbitrary tool configuration. While encouraged to use the [tool] table, PyprojectTomlConfigSettingsSource can be used to load variables from any location with in "pyproject.toml" file.

This is controlled by providing SettingsConfigDict(pyproject_toml_table_header=tuple[str, ...]) where the value is a tuple of header parts. By default, pyproject_toml_table_header=('tool', 'pydantic-settings') which will load variables from the [tool.pydantic-settings] table.

from typing import Tuple, Type

from pydantic_settings import (
    BaseSettings,
    PydanticBaseSettingsSource,
    PyprojectTomlConfigSettingsSource,
    SettingsConfigDict,
)


class Settings(BaseSettings):
    """Example loading values from the table used by default."""

    field: str

    @classmethod
    def settings_customise_sources(
        cls,
        settings_cls: Type[BaseSettings],
        init_settings: PydanticBaseSettingsSource,
        env_settings: PydanticBaseSettingsSource,
        dotenv_settings: PydanticBaseSettingsSource,
        file_secret_settings: PydanticBaseSettingsSource,
    ) -> Tuple[PydanticBaseSettingsSource, ...]:
        return (PyprojectTomlConfigSettingsSource(settings_cls),)


class SomeTableSettings(Settings):
    """Example loading values from a user defined table."""

    model_config = SettingsConfigDict(
        pyproject_toml_table_header=('tool', 'some-table')
    )


class RootSettings(Settings):
    """Example loading values from the root of a pyproject.toml file."""

    model_config = SettingsConfigDict(extra='ignore', pyproject_toml_table_header=())

This will be able to read the following "pyproject.toml" file, located in your working directory, resulting in Settings(field='default-table'), SomeTableSettings(field='some-table'), & RootSettings(field='root'):

field = "root"

[tool.pydantic-settings]
field = "default-table"

[tool.some-table]
field = "some-table"

By default, PyprojectTomlConfigSettingsSource will only look for a "pyproject.toml" in the your current working directory. However, there are two options to change this behavior.

  • SettingsConfigDict(pyproject_toml_depth=<int>) can be provided to check <int> number of directories up in the directory tree for a "pyproject.toml" if one is not found in the current working directory. By default, no parent directories are checked.
  • An explicit file path can be provided to the source when it is instantiated (e.g. PyprojectTomlConfigSettingsSource(settings_cls, Path('~/.config').resolve() / 'pyproject.toml')). If a file path is provided this way, it will be treated as absolute (no other locations are checked).
from pathlib import Path
from typing import Tuple, Type

from pydantic_settings import (
    BaseSettings,
    PydanticBaseSettingsSource,
    PyprojectTomlConfigSettingsSource,
    SettingsConfigDict,
)


class DiscoverSettings(BaseSettings):
    """Example of discovering a pyproject.toml in parent directories in not in `Path.cwd()`."""

    model_config = SettingsConfigDict(pyproject_toml_depth=2)

    @classmethod
    def settings_customise_sources(
        cls,
        settings_cls: Type[BaseSettings],
        init_settings: PydanticBaseSettingsSource,
        env_settings: PydanticBaseSettingsSource,
        dotenv_settings: PydanticBaseSettingsSource,
        file_secret_settings: PydanticBaseSettingsSource,
    ) -> Tuple[PydanticBaseSettingsSource, ...]:
        return (PyprojectTomlConfigSettingsSource(settings_cls),)


class ExplicitFilePathSettings(BaseSettings):
    """Example of explicitly providing the path to the file to load."""

    field: str

    @classmethod
    def settings_customise_sources(
        cls,
        settings_cls: Type[BaseSettings],
        init_settings: PydanticBaseSettingsSource,
        env_settings: PydanticBaseSettingsSource,
        dotenv_settings: PydanticBaseSettingsSource,
        file_secret_settings: PydanticBaseSettingsSource,
    ) -> Tuple[PydanticBaseSettingsSource, ...]:
        return (
            PyprojectTomlConfigSettingsSource(
                settings_cls, Path('~/.config').resolve() / 'pyproject.toml'
            ),
        )

Field value priority

In the case where a value is specified for the same Settings field in multiple ways, the selected value is determined as follows (in descending order of priority):

  1. If cli_parse_args is enabled, arguments passed in at the CLI.
  2. Arguments passed to the Settings class initialiser.
  3. Environment variables, e.g. my_prefix_special_function as described above.
  4. Variables loaded from a dotenv (.env) file.
  5. Variables loaded from the secrets directory.
  6. The default field values for the Settings model.

Customise settings sources

If the default order of priority doesn't match your needs, it's possible to change it by overriding the settings_customise_sources method of your Settings .

settings_customise_sources takes four callables as arguments and returns any number of callables as a tuple. In turn these callables are called to build the inputs to the fields of the settings class.

Each callable should take an instance of the settings class as its sole argument and return a dict.

Changing Priority

The order of the returned callables decides the priority of inputs; first item is the highest priority.

from typing import Tuple, Type

from pydantic import PostgresDsn

from pydantic_settings import BaseSettings, PydanticBaseSettingsSource


class Settings(BaseSettings):
    database_dsn: PostgresDsn

    @classmethod
    def settings_customise_sources(
        cls,
        settings_cls: Type[BaseSettings],
        init_settings: PydanticBaseSettingsSource,
        env_settings: PydanticBaseSettingsSource,
        dotenv_settings: PydanticBaseSettingsSource,
        file_secret_settings: PydanticBaseSettingsSource,
    ) -> Tuple[PydanticBaseSettingsSource, ...]:
        return env_settings, init_settings, file_secret_settings


print(Settings(database_dsn='postgres://postgres@localhost:5432/kwargs_db'))
#> database_dsn=MultiHostUrl('postgres://postgres@localhost:5432/kwargs_db')

By flipping env_settings and init_settings, environment variables now have precedence over __init__ kwargs.

Adding sources

As explained earlier, pydantic ships with multiples built-in settings sources. However, you may occasionally need to add your own custom sources, settings_customise_sources makes this very easy:

import json
from pathlib import Path
from typing import Any, Dict, Tuple, Type

from pydantic.fields import FieldInfo

from pydantic_settings import (
    BaseSettings,
    PydanticBaseSettingsSource,
    SettingsConfigDict,
)


class JsonConfigSettingsSource(PydanticBaseSettingsSource):
    """
    A simple settings source class that loads variables from a JSON file
    at the project's root.

    Here we happen to choose to use the `env_file_encoding` from Config
    when reading `config.json`
    """

    def get_field_value(
        self, field: FieldInfo, field_name: str
    ) -> Tuple[Any, str, bool]:
        encoding = self.config.get('env_file_encoding')
        file_content_json = json.loads(
            Path('tests/example_test_config.json').read_text(encoding)
        )
        field_value = file_content_json.get(field_name)
        return field_value, field_name, False

    def prepare_field_value(
        self, field_name: str, field: FieldInfo, value: Any, value_is_complex: bool
    ) -> Any:
        return value

    def __call__(self) -> Dict[str, Any]:
        d: Dict[str, Any] = {}

        for field_name, field in self.settings_cls.model_fields.items():
            field_value, field_key, value_is_complex = self.get_field_value(
                field, field_name
            )
            field_value = self.prepare_field_value(
                field_name, field, field_value, value_is_complex
            )
            if field_value is not None:
                d[field_key] = field_value

        return d


class Settings(BaseSettings):
    model_config = SettingsConfigDict(env_file_encoding='utf-8')

    foobar: str

    @classmethod
    def settings_customise_sources(
        cls,
        settings_cls: Type[BaseSettings],
        init_settings: PydanticBaseSettingsSource,
        env_settings: PydanticBaseSettingsSource,
        dotenv_settings: PydanticBaseSettingsSource,
        file_secret_settings: PydanticBaseSettingsSource,
    ) -> Tuple[PydanticBaseSettingsSource, ...]:
        return (
            init_settings,
            JsonConfigSettingsSource(settings_cls),
            env_settings,
            file_secret_settings,
        )


print(Settings())
#> foobar='test'

Accesing the result of previous sources

Each source of settings can access the output of the previous ones.

from typing import Any, Dict, Tuple

from pydantic.fields import FieldInfo

from pydantic_settings import PydanticBaseSettingsSource


class MyCustomSource(PydanticBaseSettingsSource):
    def get_field_value(
        self, field: FieldInfo, field_name: str
    ) -> Tuple[Any, str, bool]: ...

    def __call__(self) -> Dict[str, Any]:
        # Retrieve the aggregated settings from previous sources
        current_state = self.current_state
        current_state.get('some_setting')

        # Retrive settings from all sources individually
        # self.settings_sources_data["SettingsSourceName"]: Dict[str, Any]
        settings_sources_data = self.settings_sources_data
        settings_sources_data['SomeSettingsSource'].get('some_setting')

        # Your code here...

Removing sources

You might also want to disable a source:

from typing import Tuple, Type

from pydantic import ValidationError

from pydantic_settings import BaseSettings, PydanticBaseSettingsSource


class Settings(BaseSettings):
    my_api_key: str

    @classmethod
    def settings_customise_sources(
        cls,
        settings_cls: Type[BaseSettings],
        init_settings: PydanticBaseSettingsSource,
        env_settings: PydanticBaseSettingsSource,
        dotenv_settings: PydanticBaseSettingsSource,
        file_secret_settings: PydanticBaseSettingsSource,
    ) -> Tuple[PydanticBaseSettingsSource, ...]:
        # here we choose to ignore arguments from init_settings
        return env_settings, file_secret_settings


try:
    Settings(my_api_key='this is ignored')
except ValidationError as exc_info:
    print(exc_info)
    """
    1 validation error for Settings
    my_api_key
      Field required [type=missing, input_value={}, input_type=dict]
        For further information visit https://errors.pydantic.dev/2/v/missing
    """

In-place reloading

In case you want to reload in-place an existing setting, you can do it by using its __init__ method :

import os

from pydantic import Field

from pydantic_settings import BaseSettings


class Settings(BaseSettings):
    foo: str = Field('foo')


mutable_settings = Settings()

print(mutable_settings.foo)
#> foo

os.environ['foo'] = 'bar'
print(mutable_settings.foo)
#> foo

mutable_settings.__init__()
print(mutable_settings.foo)
#> bar

os.environ.pop('foo')
mutable_settings.__init__()
print(mutable_settings.foo)
#> foo