Acyro

English | 简体中文

A small, local, Python-native DAG task runner with deterministic execution and content-aware JSON caching.

Install

pip install acyro

Requires Python 3.12 or newer.

Quick start

from pathlib import Path

from acyro import run, task


@task(inputs=["data/*.txt"], outputs=["build/combined.txt"])
def combine() -> None:
    Path("build").mkdir(exist_ok=True)
    content = "\n".join(path.read_text() for path in Path("data").glob("*.txt"))
    Path("build/combined.txt").write_text(content)


@task(depends=[combine], outputs=["dist/report.txt"])
def report() -> None:
    Path("dist").mkdir(exist_ok=True)
    Path("dist/report.txt").write_text(Path("build/combined.txt").read_text())


run()

run() executes tasks in dependency order. Later runs mark unchanged tasks as SKIPPED.

CLI

Put tasks in a Python file, then run or inspect the graph:

acyro run examples/acyrofile.py
acyro graph examples/acyrofile.py

Cache

Successful tasks are cached under .acyro/cache. Fingerprints include task metadata, source code, input contents, output contents, and dependency fingerprints. Changing an input or dependency invalidates downstream tasks; deleting or modifying an output reruns its task.

Input declarations accept explicit files and Glob patterns. Output declarations must be explicit file paths. Paths are resolved from the current working directory, and file contents are hashed with SHA-256. Missing inputs and outputs raise clear errors without replacing the last successful cache record.

Tasks without inputs or outputs retain the original source- and dependency-based cache behavior. Use run(cache_dir=...) to choose another cache directory.

Acyro is intentionally serial and local. It has no async runtime, distributed workers, database, retries, or pluggable cache backends.

Development

uv sync --extra dev
uv run pytest -q
uv run ruff check .
uv build

Licensed under the MIT License.