<!-- BEGIN:nextjs-agent-rules -->
# This is NOT the Next.js you know

This version has breaking changes — APIs, conventions, and file structure may all differ from your training data. Read the relevant guide in `node_modules/next/dist/docs/` before writing any code. Heed deprecation notices.
<!-- END:nextjs-agent-rules -->

# Comparaison — Project Agent Rules

## LLM Provider

- **Base URL**: `https://nano-gpt.com/api/v1` (not `api.nano-gpt.com`)
- **API Key**: Use `NANOGPT_API_KEY` env var from the host's Docker `.env`
- **Model**: `minimax/minimax-m2.7` (supports `response_format: {type: json_object}`)
- **Schema tolerance**: The LLM output may deviate from the requested schema. The provider code (`src/lib/llm/providers/openai.ts`) includes `normalizeResult()` to handle common deviations (overallScore as object, sources as strings, dimensions as array of objects). Do NOT assume the LLM returns the exact schema.

## Development

```bash
# Local dev
npm run dev

# Build Docker image
docker compose build

# Run container
docker compose up -d

# Run E2E tests
npm install
npm run test:e2e:install
npm run test:e2e
```

## Deployment

Deployed to `ubuntu:/srv/compose/comparaison/` via `docker compose`.
Traefik routes `comparaison.local.tophermayor.com` → container port 3000.

## Environment Variables

Required in `.env`:
- `DATABASE_URL` — PostgreSQL connection string
- `BETTER_AUTH_SECRET` — Session secret (min 32 chars)
- `BETTER_AUTH_URL` — Auth callback URL
- `LLM_API_KEY` — NanoGPT API key (Hermes `NANOGPT_API_KEY` works)
- `LLM_BASE_URL` — `https://nano-gpt.com/api/v1`
- `LLM_MODEL` — `minimax/minimax-m2.7`

## Key Routes

| Route | Method | Auth | Description |
|-------|--------|------|-------------|
| `/api/compare` | POST | Yes | Start SSE comparison stream |
| `/api/comparisons/[slug]` | GET | Yes | Get comparison by slug |
| `/api/user/comparisons` | GET | Yes | List user's comparisons |
| `/api/user/stats` | GET | Yes | User statistics |
| `/api/auth/sign-in/email` | POST | No | Email/password sign-in |

## Testing

E2E tests live in `e2e/` and use Playwright. See `e2e/README.md` for setup.