// What powers this chatbot
// Model: Cloudflare Workers AI
// Llama 4 Scout 17B — streaming SSE
// Rate limited: 4 msg/min, 30/h per IP

const SYSTEM_PROMPT = `
  You are Claude, the AI that co-built
  pixelium.win with Stéphane Ferreira.

  Your job: answer questions about his
  skills, experience, and projects.

  Every fact is verifiable — git history,
  ops journal, or public profiles.

  61 services in production
  4 Proxmox nodes, 76 LXC
  HTB Pro Hacker #633 global
  56 Ansible playbooks, 65 hosts
  Internal PKI, SSO, IPS, SIEM, VPN
  ...
`;

const stream = await env.AI.run(
  '@cf/meta/llama-4-scout-17b-16e-instruct',
  {
    messages: [
      { role: 'system', content: SYSTEM_PROMPT },
      ...history,
    ],
    stream: true,
    max_tokens: 512,
  }
);

# wrangler.toml — Cloudflare Workers config

name = "pixelium-site"
compatibility_date = "2025-01-01"

[ai]
binding = "AI"

[[kv_namespaces]]
binding = "SESSION"    # rate limiting

[[kv_namespaces]]
binding = "STATUS_KV"  # live infra status

[[kv_namespaces]]
binding = "STATS_KV"   # portfolio stats

[[d1_databases]]
binding = "HISTORY_DB" # uptime history

// POST /api/chat — unified AI endpoint

export const POST: APIRoute = async ({ request }) => {
  // Rate limit per IP via KV
  const ip = request.headers.get('cf-connecting-ip');
  const rl = await checkRateLimit(ip);
  if (!rl.allowed) return 429;

  // mode: "cv" | "sysop"
  const { mode, messages } = await request.json();
  const prompt = PROMPTS[mode];

  // Stream response via SSE
  const stream = await env.AI.run(MODEL, {
    messages: [
      { role: 'system', content: prompt },
      ...messages.slice(-10),
    ],
    stream: true,
  });

  return new Response(stream, {
    headers: { 'Content-Type': 'text/event-stream' }
  });
};