archtool 2.0.1

Auto-wiring DI + layer enforcement for Python. Define interfaces, write implementations — archtool assembles everything.

archtool

Auto-wiring dependency injection and layer enforcement for Python.
Define interfaces. Write implementations. archtool assembles everything at startup — zero registration code, zero wiring boilerplate.

---

The problem

Every Python backend eventually grows into this:

# entrypoints/run.py
user_repo = UserRepo()
order_repo = OrderRepo()
user_service = UserService()
user_service.repo = user_repo          # don't forget this
order_service = OrderService()
order_service.repo = order_repo
order_service.user_service = user_service  # or this
# ... 40 more lines

Manual wiring doesn't scale. Every new dependency means touching the entrypoint. Every new dev breaks it.

The solution

from pathlib import Path
from archtool.dependency_injector import DependencyInjector
from archtool.global_types import AppModule

injector = DependencyInjector(
    modules_list=[
        AppModule("app.users"),
        AppModule("app.orders"),
        AppModule("app.payments"),
    ],
    project_root=Path(__file__).parent.parent,
)
injector.inject()

archtool scans your modules, discovers every interface–implementation pair, instantiates them in dependency order, and wires everything together.

---

How it works

Declare the dependency as a class annotation on the concrete class:

# app/orders/services.py
from app.users.interfaces import UserServiceABC   # cross-module dep
from .interfaces import OrderServiceABC, OrderRepoABC

class OrderService(OrderServiceABC):
    repo: OrderRepoABC        # archtool sets this
    user_svc: UserServiceABC  # and this — from another module

    def place(self, user_id: str, items: list) -> dict:
        user = self.user_svc.get(user_id)
        return {"user": user, "items": items}

Two-pass injection algorithm:

1. Pass 1 — scans interfaces.py in each module, finds abstract/concrete pairs, instantiates concretions, builds the registry
2. Pass 2 — reads class-level annotations from each instance, looks up the registry, calls setattr

No container class. No decorator. No __init__ parameters. If the class is in the right file and inherits the right base — it's wired.

---

Layer enforcement

Declare your layers and archtool enforces boundaries at startup:

injector = DependencyInjector(
    modules_list=[...],
    layers=default_layers,  # Infrastructure → Domain → Application → Presentation
)
injector.inject()  # raises TopLevelLayerUsingException if a boundary is crossed

A service depending on a controller, a domain class importing infrastructure directly — caught before they reach production.

---

Install

pip install archtool

Supports Python 3.10 · 3.11 · 3.12 · 3.13.

---

Quickstart

archtool init myapp
cd myapp
pip install -e ".[dev]"
archtool add-module orders
archtool add-module payments
python entrypoints/run.py

archtool init scaffolds a full layered-architecture project. archtool add-module creates the module files and auto-registers it — no manual edits required.

---

Documentation

0nliner.github.io/archtool

• Quickstart — working project in 5 minutes
• How it works — two-pass injection explained
• Layer enforcement — clean architecture built in
• Comparison — vs dependency-injector, injector, manual DI

---

License

MIT — Чудайкин Александр · Бюро автоматизации процессов