🏪 WarungBrain

Multi-agent AI assistant untuk UMKM yang jualan online di Tokopedia, Shopee, & WhatsApp.

Powered by Xiaomi MiMo · Built with Hermes Agent patterns

🌐 Live demo: teguh407.github.io/warungbrain

---

Kenapa WarungBrain?

UMKM Indonesia jualan online sambil ngerangkap 5 peran sekaligus: cek harga kompetitor, hitung margin, bikin caption, balesin WA customer, sampai lembur tutup buku. Hasilnya: harga ketinggalan pasar, caption seadanya, customer kabur karena slow response.

WarungBrain ngambil 4 peran itu jadi 4 AI agent yang saling ngomong. Lo input nama produk, dia kasih:

Riset kompetitor → strategi harga → konten siap-post → balesan WA siap-kirim

Satu pipeline. Dari nol jadi siap jualan dalam 30 detik.

---

✨ Fitur

• 🕵️ CompetitorScout — pull harga & rating dari Tokopedia + Shopee per produk
• 💰 PricingStrategist — kasih rekomendasi harga (3 tier: agresif / balance / premium) + margin calc
• ✍️ ContentCreator — generate caption Instagram/TikTok + deskripsi produk yang SEO-friendly
• 💬 CSAutoReply — balesan WA template untuk pertanyaan harga, stok, ongkir, refund
• 🔌 MiMo-native — pakai Xiaomi MiMo sebagai reasoning engine (drop-in OpenAI-compatible)
• 📊 Audit trace — tiap agent simpan input/output JSON, gampang debug
• 🌐 CLI + FastAPI — pake terminal atau pasang sebagai service

---

🏗️ Arsitektur

                     ┌─────────────────────┐
                     │   PRODUK INPUT      │
                     │ "kaos polos cotton" │
                     └──────────┬──────────┘
                                │
                                ▼
                     ┌─────────────────────┐
                     │ 🕵️  CompetitorScout │  scrape Tokopedia
                     │   (web tools)       │  + Shopee
                     └──────────┬──────────┘
                                │
                ┌───────────────┼───────────────┐
                ▼               ▼               ▼
      ┌──────────────┐ ┌──────────────┐ ┌──────────────┐
      │ 💰 Pricing   │ │ ✍️  Content  │ │ 💬 CSAutoReply│
      │ (MiMo logic) │ │ (MiMo write) │ │ (MiMo persona)│
      └──────┬───────┘ └──────┬───────┘ └──────┬───────┘
             └────────────────┼────────────────┘
                              ▼
                   ┌─────────────────────┐
                   │  📦 BUNDLE READY    │
                   │  • 3 tier harga     │
                   │  • 5 caption variant│
                   │  • 8 reply template │
                   └─────────────────────┘

Tiap agent stateless, ngomong via Pydantic message. CompetitorScout fan-out hasil ke 3 agent lain in parallel, output di-merge jadi satu bundle siap pakai.

---

🚀 Quickstart

1. Clone & install

git clone https://github.com/teguh407/warungbrain.git
cd warungbrain
python3.11 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. Set MiMo API key

Daftar di platform.xiaomimimo.com → ambil API key.

cp .env.example .env
# Edit .env, paste:
# MIMO_API_KEY=sk-...

3. Jalanin

CLI mode:

python -m src.main "kaos polos cotton combed 30s"

Output:

🏪 WarungBrain v0.1.0  ·  kaos polos cotton combed 30s

[1/4] 🕵️  Riset kompetitor ............ ✓ 18 listing terkumpul
[2/4] 💰 Hitung strategi harga ......... ✓ 3 tier
[3/4] ✍️  Bikin konten ................. ✓ 5 caption + 1 deskripsi
[4/4] 💬 Generate balasan CS ........... ✓ 8 reply template

📦 Bundle siap → out/bundle-20260518-1432.json

API mode (FastAPI):

uvicorn src.api:app --reload
# Buka http://localhost:8000/docs

---

📦 Output Sample

{
  "product": "kaos polos cotton combed 30s",
  "market_summary": {
    "competitor_count": 18,
    "median_price": 45000,
    "min_price": 28000,
    "max_price": 89000,
    "avg_rating": 4.7
  },
  "pricing": {
    "agresif":  { "harga": 39000, "margin_pct": 18, "rationale": "..." },
    "balance":  { "harga": 49000, "margin_pct": 35, "rationale": "..." },
    "premium":  { "harga": 69000, "margin_pct": 55, "rationale": "..." }
  },
  "content": {
    "captions_ig": ["caption 1", "caption 2", "..."],
    "captions_tiktok": ["...", "..."],
    "deskripsi_produk": "..."
  },
  "cs_replies": {
    "tanya_harga":  "Halo Kak! Untuk kaos polos cotton combed 30s...",
    "tanya_stok":   "...",
    "tanya_ongkir": "...",
    "complain_size":"...",
    "refund":       "..."
  }
}

---

🧠 Stack

Python 3.11 + Pydantic     → schemas + validation
Xiaomi MiMo (mimo-pro)     → reasoning (pricing, content, CS)
httpx + BeautifulSoup      → competitor scraping
DuckDuckGo Search          → fallback site: search untuk Tokped/Shopee
FastAPI                    → REST API
Rich CLI                   → progress + output

---

🛠️ Konfigurasi

Semua knob ada di .env:

MIMO_API_KEY=sk-...
MIMO_MODEL=mimo-pro
MIMO_BASE_URL=https://platform.xiaomimimo.com/v1

ENABLE_TOKOPEDIA=true
ENABLE_SHOPEE=true
MAX_LISTINGS_PER_SOURCE=10

DEFAULT_HPP_RATIO=0.4    # asumsi default cost = 40% dari median market
OUTPUT_DIR=./out

Lo bisa override DEFAULT_HPP_RATIO per-product kalau punya data HPP real.

---

🗺️ Roadmap

• 4-agent pipeline
• Tokopedia + Shopee scout (graceful degradation kalau anti-bot)
• 3-tier pricing strategy
• Caption IG + TikTok + deskripsi produk
• CS auto-reply 8 template
• FastAPI server
• WhatsApp bot integration (Whapi / Baileys)
• Ongkir kalkulator (RajaOngkir API)
• Auto-post scheduler (Instagram Graph API)
• Dashboard React untuk multi-produk
• Marketplace listing auto-publish

---

⚠️ Catatan Scraping

Tokopedia & Shopee aktif anti-bot. Default scout pakai dua jalur:

1. DuckDuckGo site:tokopedia.com search — handal headless, no proxy needed
2. Direct fetch — fallback ke API publik mereka, bisa rate-limited

Buat production scale, tambahin Playwright + residential proxy. Lihat src/tools/scout_backends.py buat extension point.

---

🤝 Contributing

PR welcome. Run pytest sebelum buka PR. Style: ruff format.

---

📜 License

MIT — see LICENSE.

---

🙏 Acknowledgements

• Xiaomi MiMo — reasoning engine yang nyalain semua agent
• Hermes Agent — multi-agent pattern + skill system
• UMKM Indonesia — yang udah jadi backbone ekonomi & inspirasi project ini 🇮🇩