yeetr, build tiny CLIs. Easy to code. Based on Python type hints.

yeetr

A tiny, typed, signature-driven CLI runner.

PyPI distribution: yeetr Python import package: yeetr CLI command: yeet

No decorators. No command classes. No ceremony. Just yeet the function.

Minimal example

Zero-boilerplate: the yeet script

Installing yeetr also installs a yeet script that finds and runs a function in any Python file.

No if __name__ == "__main__" block, no yeetr.run(...) call — just the function:

# app.py
def main(thing: int, *, n: float = 0.1) -> None:
    print(thing, n)

yeet file.py 5 --n 0.2

The default function name is main. Pass a different one to pick another top-level function in the same file:

# app.py
def main(...) -> None: ...
def greet(name: str, *, loud: bool = False) -> None: ...

yeet file.py greet world --loud

yeet file.py --help prints the target function's help, not yeet's. yeet itself only has yeet FILE [FUNC] [args...].

You can still use the explicit yeetr.run(main) form when you prefer — the yeet script is just sugar on top of it.

Explicit yeetr.run(main)

def main(thing: int, *, n: float = 0.1) -> None:
    print(thing, n)


if __name__ == "__main__":
    import yeetr
    yeetr.run(main)

yeet file.py 5 --n 0.2

Note the bare * in the signature: parameters before it become positional CLI args, parameters after it become --options. That's the whole mapping — no decorators, no per-parameter annotations needed.

Hashbang

For tiny scripts, you can make the file itself executable and let yeet discover main directly from the shebang. The short forms are:

#!yeet

or:

#!uv run yeet

For example:

#!yeet

def main(name: str, *, loud: bool = False) -> None:
    print(name.upper() if loud else name)

Then run it directly:

chmod +x greet.py
./greet.py world --loud

If you need a different entry function, keep the shebang simple and call uv run yeet file.py other_func ... explicitly instead.

Async

async def main(name: str, *, loud: bool = False) -> None:
    ...

yeet file.py world --loud

If the function is a coroutine, its result is awaited via asyncio.run, or via uvloop.run when the optional uvloop extra is installed:

uv add "yeetr[uvloop]"

When uvloop is importable, yeetr uses it transparently — no code change required. Otherwise it falls back to the stdlib event loop.

Path

from pathlib import Path


def main(path: Path, *, output: Path | None = None) -> None:
    ...

yeet file.py input.pdf --output out.txt

Literal choices

from typing import Literal


def main(*, format: Literal["json", "csv"] = "json") -> None:
    ...

yeet file.py --format csv

Arg and Opt metadata

For aliases and help text, use Arg (positional) or Opt (keyword-only) inside Annotated:

from pathlib import Path
from typing import Annotated
from yeetr import Arg, Opt


def main(
    path: Annotated[Path, Arg(help="Input file")],
    *,
    workers: Annotated[int, Opt(alias="-w", help="Worker count")] = 4,
) -> None:
    ...

yeet file.py input.pdf -w 8

Arg accepts help, metavar, min, and the path validators below. Opt accepts alias, aliases, help, metavar, envvar, hidden, and the path validators below. Mixing them (e.g. Opt on a positional or Arg on a keyword-only parameter) raises a clear YeetrError.

You can also define aliases once and reuse them:

from pathlib import Path
from typing import Annotated
from yeetr import Arg, Opt


type InputPath = Annotated[Path, Arg(help="Input file")]
type WorkerCount = Annotated[int, Opt(alias="-w", help="Worker count")]


def main(path: InputPath, *, workers: WorkerCount = 4) -> None:
    ...

Environment variable fallback (Opt(envvar=...))

Opt(envvar="NAME") falls back to an environment variable when the flag is not provided on the CLI. Precedence: explicit CLI > env var > default.

from typing import Annotated
from yeetr import Opt


def main(*, workers: Annotated[int, Opt(envvar="WORKERS")] = 4) -> None:
    ...

WORKERS=8 yeet file.py        # workers == 8
yeet file.py --workers 16     # workers == 16 (CLI wins)
yeet file.py                  # workers == 4  (default)

Env-var values are type-coerced just like CLI values. bool accepts 1/0/true/false/yes/no (case-insensitive). list[T] splits on os.pathsep (: on POSIX, ; on Windows). Literal choices are validated.

Hidden options (Opt(hidden=True))

Hidden options still parse from the CLI but are absent from --help (both the usage line and the options table):

from typing import Annotated
from yeetr import Opt


def main(*, debug: Annotated[bool, Opt(hidden=True)] = False) -> None:
    ...

Path validators

Arg and Opt accept exists, file_okay, dir_okay, readable, and writable for Path parameters. They run at parse time and fail with a clear error:

from pathlib import Path
from typing import Annotated
from yeetr import Arg


def main(
    src: Annotated[Path, Arg(exists=True, dir_okay=False, readable=True)],
    dst: Annotated[Path, Arg(writable=True)],
) -> None:
    ...

Defaults mirror typer: file_okay=True, dir_okay=True, others off. Setting any path-check on a non-Path parameter raises YeetrError at parser-build time. Validators also apply to list[Path] and to *paths: Path.

Variadic positional args (*args)

*args maps to a trailing variadic positional CLI argument. The annotation on *args is the element type (not list[T]):

from pathlib import Path


def main(dst: Path, *sources: Path) -> None:
    ...

yeet file.py dst src1 src2 src3

By default *args accepts zero or more values (argparse nargs="*"). Use Arg(min=1) to require at least one:

from typing import Annotated
from yeetr import Arg


def main(*sources: Annotated[Path, Arg(min=1, help="Source paths")]) -> None:
    ...

Keyword-only options remain --flags after *args. **kwargs is not supported.

Why Annotated? Python's type system only permits call expressions (Opt(...)) inside the metadata slot of Annotated. No other syntax is accepted by Pyright in strict mode. The Annotated form is verbose but is the only way to attach per-parameter metadata that fully type-checks.

Rules

• Positional parameters become positional CLI args.
• Keyword-only parameters (after *) become --options.
• Names convert from snake_case to kebab-case for CLI flags.
• flag: bool = False becomes --flag.
• flag: bool = True becomes --no-flag.
• Required bool parameters raise a clear error.
• T | None / Optional[T] are accepted; treated as their inner type with None as default.
• list[T] becomes a repeated option (--tag a --tag b).

Supported types

str, int, float, bool, pathlib.Path, typing.Literal[...], T | None, list[T]. Anything else raises a clear YeetrError.

Logging

By default, yeetr.run installs a Rich-based logging handler before invoking your function, so you get formatted logs with zero boilerplate:

import logging

import yeetr

logger = logging.getLogger("app")


def main(thing: int) -> None:
    logger.info("thing = %s", thing)

If your function has a log_level parameter (e.g. log_level: Literal["debug", "info", "warning", "error"] = "info"), its value drives the log level. Otherwise, the default is INFO.

Setup is idempotent: if the root logger already has handlers, yeetr does not touch them. To take full control of logging yourself, opt out:

yeetr.run(main, should_setup_logging=False)

Testing

run() accepts an explicit argv for tests:

yeetr.run(main, argv=["5", "--n", "0.2"])

yeetr vs. typer

Typer is a mature, feature-rich CLI framework and a direct inspiration for yeetr — the Annotated[..., Arg/Opt] metadata pattern, path validators, and envvar fallback all take cues from typer. yeetr is a much smaller library aimed at a narrower slice of the problem. Quick honest comparison so you can pick the right tool:

If you need subcommands or shell completion, use typer. If you want one function = one CLI with minimal ceremony and strict typing, yeetr is designed for that.

Releases

yeetr uses CalVer based on the release date. Versions are published in PEP 440 canonical form as YYYY.M.D, so a release on 2026-05-21 is 2026.5.21; multiple releases on the same day use .postN, for example 2026.5.21.post1.

Run task release to create the release/{TAG} PR, then merge it. Then create and push the matching release tag. GitHub Actions validates the tag, creates the GitHub Release, and a separate workflow deploys docs.

If you need to bypass the PR flow, run task release-direct. That bumps the version on main, runs task deps-lock, commits, pushes main, creates the matching tag, and pushes the tag.

To bump a release version manually, run uv version <version>.

Install from PyPI with:

pip install yeetr

Development

uv sync
uv run ruff check
uv run pyright
uv run pytest