Skip to content



The Python standard library os module offers operating system utilities, including os.environ for working with environment variables. os.environ only handles strings, and doesn't include tools for file I/O.

os.environ is instantiated from os._Environ, which is an implementation of The source code for os.environ can be found in python/cpython/Lib/, and the source code for can be found in python/cpython/Lib/ As explained in the docstring for MutableMapping:

A MutableMapping is a generic container for associating key/value pairs.

This class provides concrete generic implementations of all methods except for __getitem__, __setitem__, __delitem__, __iter__, and __len__.

Subclasses of, such as os._Environ, need to provide implementations of each of those methods.

Type stubs for os.environ are in python/typeshed/stdlib/os/__init__.pyi. Stubs for are in python/typeshed/stdlib/typing.pyi (these are the type stubs for the standard library typing module, which aliases

See the os module docs and the docs on mapping types for more information.


  • environs is a project with useful features for managing .env files and application settings. Its API was inspired by django-environ and envparse (and the maintainers considered merging environs and envparse into one project). The README compares environs to os.environ, and justifies the additional features that environs provides.
  • While it separates config from code, as suggested by the twelve-factor app methodology, it also combines environment variables and settings. Environment variables and their type-casted setting counterparts are combined into the same model.
  • Initially, the source code wasn't consistently type-annotated (sloria/environs#186), but based on its PEP 561 marker file, it appears to be type-annotated now.
  • Depends on python-dotenv (sloria/environs#196), so it inherits the limitations described in the python-dotenv section.


Settings configuration

pydantic offers a BaseSettings model. Settings class attributes are automatically read from environment variables, and the full power of pydantic data parsing/validation can be applied.

Simple pydantic settings model

import os

from pydantic import BaseSettings

os.environ["BOOLEAN_SETTING"] = "false"
os.environ["INTEGER_SETTING"] = "123"
os.environ["STRING_SETTING"] = "example_value"

class SimpleSettings(BaseSettings):
    boolean_setting: bool = True
    integer_setting: int = 000
    string_setting: str = "default_value"

# {"boolean_setting": False, "integer_setting": 123, "string_setting": "example_value"}

File I/O

  • In addition to reading environment variables that have already been set, pydantic can load environment variables from .env files. However, it depends on python-dotenv to load .env files, so it inherits the limitations described in the python-dotenv section.
  • If no .env file is found at the path provided, pydantic will fail silently, rather than raising a FileNotFoundError. This can lead to issues if applications depend on environment variables that pydantic fails to load.


  • python-decouple loads settings from .env and .ini files. Its supported configuration file format appears to be inspired by Foreman, a Ruby configuration management tool.
  • Variables are set with calls to instances of its AutoConfig class, which offers type-casting to convert strings to other Python types: config("DEBUG", cast=bool).
  • Source code is not type-annotated.
  • Classes inherit from object, and therefore require their own implementations of methods already present in other data structures. This could be easily eliminated by inheriting from a mapping data structure such as
  • Continues supporting Python 2 after its end-of-life, and has not been tested on the latest versions of Python 3.


python-dotenv is a package for loading .env files and setting environment variables. It was started by Django co-creator Jacob Kaplan-Moss in 2013, and was originally called django-dotenv. It is used by Uvicorn and pydantic, and suggested in the FastAPI docs.

Environment variables

  • Its primary data structure, dotenv.main.DotEnv, inherits from object. As a result, it requires its own mapping methods (such as dict()) that could be obviated by inheriting from a mapping data structure such as
  • Other methods have confusing, counter-intuitive APIs. For example, the load_dotenv() function is supposed to "Parse a .env file and then load all the variables found as environment variables," according to its docstring. However, the function always returns True, even if no .env file is found or no environment variables are set, because of DotEnv.set_as_environment_variables(). Furthermore, this confusing behavior is not documented, because, as the maintainer commented, "The return value of load_dotenv is undocumented as I was planning to do something useful with it, but could not settle down to one."

File I/O

  • Loads files with the synchronous open() built-in function. Async support is not provided.
  • Does not integrate with object storage like AWS S3.

Project maintenance

Comparing fastenv and python-dotenv


  • Both fastenv and python-dotenv provide a DotEnv class for managing environment variables
  • fastenv.DotEnv inherits from, dotenv.main.DotEnv inherits from object
  • fastenv includes DotEnv in its __all__, python-dotenv does not (it must be directly imported from dotenv.main)


  • fastenv: await fastenv.find_dotenv() (async)
  • python-dotenv: dotenv.find_dotenv() (sync)
  • Both fastenv and python-dotenv look for ".env" by default
  • Both python-dotenv and fastenv return os.PathLike objects
  • fastenv raises FileNotFoundError exceptions by default if files are not found, python-dotenv does not


  • fastenv: await fastenv.load_dotenv() (async)
  • python-dotenv: dotenv.load_dotenv() (sync)
  • fastenv.load_dotenv can load multiple .env files in a single call, dotenv.load_dotenv cannot
  • fastenv.load_dotenv logs the number of environment variables loaded, dotenv.load_dotenv does not
  • fastenv.load_dotenv returns a DotEnv model, dotenv.load_dotenv returns True (even if no .env file was found and no environment variables were loaded)

find_dotenv with load_dotenv

Users who would like to ensure their .env files are found, and log the result, should be aware that dotenv.load_dotenv:

  • Only calls find_dotenv if a file path is not provided, and does not pass an argument through to find_dotenv to raise exceptions if the file is not found
  • Requires a call to DotEnv.set_as_environment_variables to actually set environment variables
  • Does not provide logging
  • Does not provide exception handling (its verbose argument does not necessarily raise an exception)
  • Does not return the DotEnv instance created by load_dotenv, but always returns True, even if no .env file is found or no environment variables are set

Something like the following is therefore needed instead of using dotenv.load_dotenv:

Finding and loading a .env file with python-dotenv

import logging

from dotenv import find_dotenv
from dotenv.main import DotEnv

def find_and_load_my_dotenv(env_file: str = ".env") -> DotEnv:
        logger = logging.getLogger()
        source = find_dotenv(filename=env_file, raise_error_if_not_found=True)
        dotenv = DotEnv(source, verbose=True)
          f"Python-dotenv loaded {len(dotenv.dict())} variables from {env_file}"
        return dotenv
    except Exception as e:
        logger.error(f"Error loading {env_file}: {e.__class__.__qualname__} {e}")

The above effect can be accomplished with fastenv in a single call, await fastenv.load_dotenv(find_source=True). This call to fastenv.load_dotenv:

  • Finds the .env file (find_source=True) with its find_dotenv method and the file name provided (".env" by default), logging and raising a FileNotFoundError if not found
  • Sets environment variables automatically
  • Logs successes and errors automatically
  • Raises exceptions by default
  • Returns a DotEnv instance


  • fastenv: await fastenv.dotenv_values() (async)
  • python-dotenv: dotenv.dotenv_values() (sync)
  • fastenv.dotenv_values offers a find_dotenv argument to find files before loading and returning values, dotenv.dotenv_values does not
  • fastenv.dotenv_values offers a raise_exceptions argument to determine whether or not exceptions will be raised, dotenv.dotenv_values does not (its verbose argument does not necessarily raise an exception)
  • fastenv.dotenv_values logs successes and errors automatically, dotenv.dotenv_values does not

Writing to .env files

  • fastenv: await fastenv.dump_dotenv() (async, and writes an entire DotEnv model to a file)
  • python-dotenv: dotenv.get_key(), dotenv.set_key(), dotenv.unset_key() (sync, and can only write single variables to a file)


Settings configuration

Starlette offers a config module for working with environment variables and settings, which takes inspiration from python-decouple. Settings are created by calling a Starlette Config instance. Constant notation is suggested for settings (UPPERCASE_WITH_UNDERSCORES).


import starlette.config

config = starlette.config.Config()
FOO_CONSTANT = config("FOO_VARIABLE", cast=str, default="baz")

Starlette will look for an environment variable FOO_VARIABLE. For example, setting FOO_VARIABLE=bar in the environment will result in FOO_CONSTANT = "bar". If the environment variable is unset, the result will be FOO_CONSTANT = "baz". The cast keyword argument allows use of Starlette type-casting.


Type-casting provides improvements over some aspects of the standard library. For example, Starlette intuitively casts Boolean values, unlike the Python built-in bool type:

Type-casting with starlette.config

.venv  python3

>>> bool("false")
>>> import starlette.config
>>> config = starlette.config.Config()
>>> config("BOOLEAN_SETTING", cast=bool, default="false")

One-way configuration preference

Starlette has an opinionated one-way configuration preference (environment variables -> Starlette Config instance). To avoid modifying environment variables after they have been loaded into a Starlette Config instance, Starlette provides its own mapping onto os.environ (starlette.config.environ), which will raise an exception on attempts to change an environment variable that has already been loaded into a corresponding setting on a Config instance.

While it is useful to have starlette.config.environ synchronized with os.environ, the downside is that starlette.config.environ contains local environment variables loaded from os.environ, and therefore wouldn't typically be dumped to a file.

It is also important to note that the one-way preference will only be enforced when using starlette.environ. If a variable is changed using os.environ, it will be updated correspondingly in starlette.environ, but no exception will be raised, and the Config instance value will not be updated.

Environment variables with starlette.config

.venv  python3

>>> import os
>>> import starlette.config
>>> os.environ.get("FOO_VARIABLE")
>>> starlette.config.environ["FOO_VARIABLE"] = "bar"
>>> os.environ.get("FOO_VARIABLE")
>>> config = starlette.config.Config()
>>> FOO_CONSTANT = config("FOO_VARIABLE", cast=str, default="baz")
>>> starlette.config.environ["FOO_VARIABLE"] = "foo"
Traceback (most recent call last):
File ".venv/lib/python3.9/site-packages/starlette/", line 26, in __setitem__
    raise EnvironError(
starlette.config.EnvironError: Attempting to set environ['FOO_VARIABLE'],
but the value has already been read.
>>> os.environ["FOO_VARIABLE"] = "foo"
>>> os.environ["FOO_VARIABLE"]
>>> starlette.config.environ["FOO_VARIABLE"]

File I/O

  • Starlette Config accepts an env_file keyword argument, which should point to a .env file on disk. It loads the file with the synchronous open() built-in function.
  • If no .env file is found at the path provided by Config(env_file), Starlette will fail silently, rather than raising a FileNotFoundError. This can lead to issues if applications depend on environment variables that Starlette fails to load.
  • Starlette Config does not support multiple .env files (encode/starlette#432).

The future of starlette.config

From encode/starlette#432:

I guess we may actually end up pulling the configuration stuff out into a strictly separate package (or indeed even just pointing at python-decouple - since it's the same style).

With 12-factor config you really should have a fairly small set of environment. (Eg. the database configuration should just be a single URL.)

What we will want to do though is provide really good examples of how to arrange things sensibly, so that eg. you have a project that has a that includes a whole bunch of project configuration, and which you can arrange into separate bits, and demonstrate clearly how the environment should be a small set of variables, but the project settings may be more comprehensive.

We could perfectly well also point to other configuration-management packages that're out there as alternatives.


  • dotenvy can load .env files and apply a type schema to the variables. It does not appear to be actively maintained.
  • env does not add much beyond os.environ (does not even load files), has not been released since 2012, and does not appear to be actively maintained.
  • envparse offers features for parsing and type-casting environment variables, but does not appear to be actively maintained.
  • python-configurator depends on python-dotenv and appears to emphasize TOML settings files.