docs/multi-user-architecture.md

# Multi-User Architecture

Mplus Studio can run in one hosted Docker environment for multiple users, but Kaggle auth must be isolated per user.

## Services

- App: Angular frontend served by the Node backend.
- PostgreSQL: users, sessions, jobs, ownership, metrics, and audit records.
- MinIO: S3-compatible object storage for uploaded images, generated outputs, and reports.
- Kaggle CLI: executed server-side only, scoped to the current user's auth context.

## User Flow

Product language should stay simple:

```text
Productafbeeldingen maken
Inloggen en starten
Je kunt nu productafbeeldingen maken
```

The regular user flow should not mention Kaggle, GPU, kernel, notebook, provider, S3, or database.

## Kaggle Isolation

Each user needs one isolated Kaggle auth context.

For CLI-based auth, use a per-user config directory:

```text
/app/data/kaggle-sessions/<user-id-or-session-id>/
```

Every Kaggle command for that user must run with:

```text
KAGGLE_CONFIG_DIR=/app/data/kaggle-sessions/<user-id-or-session-id>
```

Never expose Kaggle auth files to the browser.

## Storage Model

PostgreSQL stores metadata and ownership. MinIO stores binary assets.

Suggested object keys:

```text
users/<user-id>/jobs/<job-id>/inputs/<file-name>
users/<user-id>/jobs/<job-id>/outputs/generated.png
users/<user-id>/jobs/<job-id>/report/result.md
```

## Next Implementation Step

The current MVP still generates directly from the request payload. The next backend step is to add:

- session/user table;
- job table;
- S3 upload/download helpers;
- async job status endpoint;
- per-user Kaggle config resolution.