🦉 owl · check

Your Python service must not start with a broken environment. owlcheck enforces that.

One call at the top of your application's entrypoint validates that every environment variable your code declares it needs is present and well-typed. The process either starts with a complete configuration or it does not start at all.

This eliminates a specific class of production failure: the service that boots cleanly, accepts traffic for ten minutes, then crashes the first time a code path reaches for os.environ.get("STRIPE_SECRET"). With owlcheck, that variable is checked before the first request — or the process exits before any request can arrive.

Status

Early development. API is not yet stable. Requires pydantic>=2, pydantic-settings>=2, Python 3.10+.

Installation

pip install owlcheck

The canonical pattern

There is exactly one way to use owlcheck. This is by design.

# main.py
from owlcheck import Settings
from pydantic import Field


class AppSettings(Settings):
    database_url: str = Field(..., alias="DATABASE_URL", description="Postgres DSN")
    stripe_secret: str = Field(..., alias="STRIPE_SECRET", description="Stripe API secret")
    debug: bool = Field(False, alias="DEBUG")


settings = AppSettings.load_or_exit()

That's the entire integration. After this line returns, settings is a validated, immutable record of your application's configuration. Import it anywhere:

from main import settings

connect(settings.database_url)

If any required environment variable is missing or invalid, the process exits with code 1 and a log message that names the violated contract — before any handler, worker, or background task can run.

What this enforces

• No partial-config startup. Validation runs before your application accepts any work. If something required is missing, the process fails immediately with a readable report; it does not run for ten minutes and crash at request time.
• One source of truth. Your Settings subclass is the schema. Scattered os.environ.get(...) calls in handlers, workers, and helpers are explicitly the failure mode owlcheck exists to prevent.
• Aliased error messages. Validation errors name the environment variable (the field's alias), not the Python attribute. Operators read DATABASE_URL: missing (Postgres DSN), not database_url: field required.
• Undeclared .env keys are surfaced. Any variable in your .env file that isn't a declared field is reported at startup - that undeclared surface is the failure mode owlcheck exists to remove - and ones that look like a typo of a field name get a "did you mean" hint. (Ambient environment variables aren't pulled in; only keys actually in the file.) Every extra is then dropped so nothing undeclared reaches your code.
• Per-field status log. Each declared field is reported on startup as provided (from env or override) or default. You always know what configuration the process actually loaded.
• Immutable after load. Once load_or_exit() returns, the contract is frozen. Attempts to mutate settings.x = ... raise ValidationError. The configuration of a running process does not change under the code that depends on it.

Why "one call"

The alternative — scattered os.getenv("X") checks across handlers, workers, and helpers — is a tax that compounds:

1. A contributor adds a feature that reads os.getenv("NEW_FLAG"). Code review does not catch it; nothing connects NEW_FLAG to the declared schema.
2. The feature ships. In production, NEW_FLAG is unset. The feature degrades silently in some flows, loudly in others. Nobody notices until a customer complains.
3. You spend an afternoon greping for every os.getenv and os.environ.get in the repo to figure out what the actual configuration surface of the service is.

owlcheck makes one canonical pattern strictly easier than every alternative. Declare every environment variable your service needs as a Field on your Settings subclass. Call load_or_exit() once in main.py. Use settings.x everywhere else. The schema is the contract; the call enforces it.

Public API

Design choices

• Honest typing only. If a field can be None at runtime, declare it T | None. owlcheck does not provide "soft optional" markers that lie to the type checker.
• Library, not framework. owlcheck does not configure logging for you. Configure your application's logging once; owlcheck writes to logging.getLogger("owlcheck") like any well-behaved library.
• Immutable after load. Once load() returns, the contract is frozen. See above.

License

MIT