Architecture preset / feature matrix

Interactive fastkit init --interactive asks for an architecture preset (issue #44) before collecting feature selections. The preset shapes the generated project's layout — different presets ship a different base template and put generated config files in different locations so they sit next to the existing structure rather than in a parallel src/config/ tree.

This page is the source of truth for what each preset does, where files land, and which feature combinations require manual wiring.

Preset → base template

Preset	Base template	Description
`minimal`	`fastapi-empty`	Smallest viable FastAPI app — placeholder `main.py` is regenerated from your feature selections.
`single-module`	`fastapi-single-module`	Single-file FastAPI app — `main.py` is regenerated.
`classic-layered`	`fastapi-default`	Layered split (`api/routes`, `crud`, `schemas`, `core`). Shipped `main.py` is preserved.
`domain-starter`	`fastapi-domain-starter`	Domain-oriented (`src/app/domains/<concept>/`). Shipped `main.py` is preserved. Recommended default.

Generated file locations

Preset	`main.py` overlay	Database config target	Auth config target
`minimal`	regenerated at `src/main.py`	`src/config/database.py`	`src/config/auth.py`
`single-module`	regenerated at `src/main.py`	`src/config/database.py`	`src/config/auth.py`
`classic-layered`	preserved (template-shipped)	`src/core/database.py`	`src/core/auth.py`
`domain-starter`	preserved (template-shipped)	`src/app/core/database.py`	`src/app/core/auth.py`

Database / auth feature support per preset

These features are supported across every preset — the package install always succeeds; the difference is whether the dynamic main.py overlay also wires them up automatically.

Feature	`minimal` / `single-module`	`classic-layered` / `domain-starter`
Database (PostgreSQL, MySQL, SQLite, MongoDB)	Generates the config module and stubs `await init_db()` calls in the regenerated `main.py`.	Generates the config module at the preset's path. The shipped `main.py` is preserved, so wire `get_db()` into routers manually.
Authentication (JWT, FastAPI-Users, OAuth2, Session-based)	Generates the auth config module. JWT also imports `HTTPBearer` in the regenerated `main.py`.	Generates the auth config module at the preset's path. No imports added to `main.py` — wire dependencies manually.
Background tasks (Celery, Dramatiq)	Packages installed; no main.py overlay today.	Same.
Caching (Redis)	Packages installed; no main.py overlay today.	Same.
CORS (utility)	`CORSMiddleware` added to the regenerated `main.py` with `allow_origins=['*']`.	Already wired in the shipped `main.py` (conditional on `settings.all_cors_origins`). Activate by setting `BACKEND_CORS_ORIGINS` in `.env` — no code edits required.
Testing (Basic / Coverage / Advanced)	`pytest.ini` is generated at the project root.	Same.
Deployment (Docker, docker-compose)	`Dockerfile` and/or `docker-compose.yml` written at the project root.	Same.

When you'll see a "Preset compatibility" warning

For presets that preserve the shipped main.py (classic-layered, domain-starter), some feature selections won't be auto-wired into the app. The CLI surfaces a one-shot warning at the end of generation listing which selections need manual wiring:

Selected feature	Triggers a warning under `classic-layered` / `domain-starter`?
`CORS` (utility)	❌ — already wired in the shipped `main.py`. Just populate `BACKEND_CORS_ORIGINS` in `.env`.
`Rate-Limiting` (utility)	✅ — `slowapi` limiter setup is not added
`Prometheus` (monitoring)	✅ — `Instrumentator().instrument(app)` is not called
Any database / auth selection	⚠️ — config files are generated, but you must `Depends()` them into your routers

For minimal and single-module presets the dynamic main.py overlay handles CORS, rate-limiting, and Prometheus instrumentation automatically; no warnings fire.

Unsupported combinations (stay safe)

The strategist deliberately does not attempt to splice generated code into a template-shipped main.py. Doing so would risk producing broken imports or duplicating routers. The contract is:

Selected packages are always installed (so pip freeze matches the user's intent).
Generated config modules always land at the preset-appropriate path.
For preserve-main presets, the user is told which selections still need manual wiring instead of getting silently broken code.

If you need full auto-wiring of every feature, pick minimal or single-module — they regenerate main.py from feature flags.