Overview · dazo/docs

engine-bot (package: chatbot-dazo) adalah AI agent service ekosistem Dazo — backend chatbot e-commerce multi-tenant untuk percakapan WhatsApp. Service ini menangani seluruh lifecycle percakapan: pencarian produk, cart, checkout, ongkir, pembayaran, bukti transfer, dan follow-up terjadwal melalui arsitektur multi-agent LangChain.

Peran di ekosistem

Aspek	Penjelasan
Identitas	Package `chatbot-dazo`, repo `engine-bot`
Posisi	Antara `backend-go` (pemanggil) dan OpenAI/MongoDB/DazoApp/DOKU
Tenant	`store_id` — di-scope via query aplikasi, bukan database terpisah
Datastore	MongoDB primer (DB `dazo` lokal / `dazodev` remote)
Queue	Redis + BullMQ — delayed follow-up jobs
Auth	JWT HS256 — validasi-only, issued by dazoapp
AI	LangChain/LangGraph + OpenAI — hierarchical multi-agent
Realtime	Socket.IO — event `newMessage`

Arsitektur

text

backend-go (:8081) / WA client
    │ POST /api/chatbot-api
    ▼
Express (:1313)  ←→  MongoDB (dazo)
    │                    ▲
    ├─ verifyToken        │
    ├─ mainController     │ domain services + tools
    ├─ context loader     │
    ├─ main orchestrator  │
    ├─ specialist agent   │
    ├─ domain tools       │
    ▼                     │
OpenAI (gpt-4o / gpt-5.4) ┘
    │
    ├─ Redis + BullMQ (follow-up queue)
    ▼
Worker (scheduler/worker/worker.js)
    │
    ▼
DazoApp / DOKU / RajaOngkir / Messaging API

Engine Bot memakai hierarchical multi-agent architecture. HTTP controller menyiapkan context tenant dan customer, main orchestrator memilih specialist agent, lalu specialist menjalankan LangChain tools yang membaca atau mengubah MongoDB dan API eksternal.

Stack teknologi

Lapisan	Teknologi
Bahasa	Node.js ES modules
Web framework	Express 4.21
Database	MongoDB via Mongoose 8.7
AI orchestration	LangChain 1.2 + LangGraph 1.1
LLM	OpenAI (`gpt-4o`, `gpt-5.4`, `gpt-4o-mini`, `gpt-4o-transcribe`)
Vector DB	FAISS (`faiss-node`)
Queue	Redis + BullMQ 5.70
Auth	JWT (`jsonwebtoken`)
Realtime	Socket.IO 4.8
Ongkir/tracking	RajaOngkir / Komerce
Instant payment	DazoApp → DOKU
NLP	`node-nlp` / `@nlpjs/lang-id` (intent detection)
Cron	`node-cron`
Deploy	GitHub Actions → PM2

Komponen utama

Komponen	Lokasi	Catatan
HTTP server	`app.js` (60 baris)	Express, CORS, Socket.IO, MongoDB startup, Redis health, routes
Routes	`routes/chatbotRoutes.js` (35 baris)	Semua endpoint `/api`, JWT middleware
Main controller	`controllers/mainController.js`	Buffer, dedup, context, orchestration, response, scheduling
Main prompt	`prompt/main-agent.md`	Aturan routing ke specialist agent
Order agents	`subagents/subagents-orders.js`	Customer service, admin, logistics
Payment agent	`subagents/subagents-payment.js`	Metode bayar, konfirmasi, status instant payment
Booking agent	`subagents/subagents-booking.js`	Facade booking; tool backend belum diimplementasikan
Order tools	`tools/order/tools.js`	Cart, checkout, address, shipping, payment, complaint
Domain services	`service/`	Query, formatting, order lifecycle, external API
Models	`models/` (24 file)	Mongoose schema
Follow-up producer	`scheduler/workerService.js`	Generate dan enqueue delayed job
Follow-up worker	`scheduler/worker/worker.js`	Re-check order dan kirim pesan WhatsApp
Middleware	`middleware/authMiddleware.js`	`verifyToken`
Config	`config/env.js`, `config/db.js`	Environment + Mongoose connect
Utils	`utils/util.js`, `utils/redisHealth.js`	Vision, transcription, Redis health

Specialist agents

Agent	Tool wrapper	Model	Fungsi
Customer Service	`customer_service_department`	`gpt-4o-mini`	FAQ, katalog, stok, komplain
Admin	`admin_department`	`gpt-5.4`	Cart, checkout, perubahan produk/varian/quantity
Logistics	`logistics_departemen`	`gpt-5.4`	Alamat, ongkir, kurir, tracking resi
Payment	`payment_department`	`gpt-5.4`	Metode bayar, konfirmasi, instant payment
Booking	`booking_departement`	`gpt-4o-mini`	Reservasi event — tool masih stub

Cara kerja singkat

Inbound: backend-go memanggil POST /api/chatbot-api dengan store_id, customer_id, agent_id, dan pesan. Controller menyiapkan context tenant, deteksi intent, lalu main orchestrator memanggil satu specialist tool berdasarkan prompt.

Follow-up: Setelah percakapan tertentu, scheduler mengenqueue delayed job ke Redis/BullMQ. Worker memeriksa ulang status order/payment sebelum mengirim pesan WhatsApp terjadwal.

Detail alur ada di Agent Architecture dan Order Flow.

Langkah berikutnya

Baru mulai? Baca Setup Environment — config/env.js memegang seluruh URL service dan kredensial.
Sudah punya env? Lanjut ke Local Development.
Akan deploy? Baca Deployment — PM2, dua process (app + worker).
Engineer baru wajib baca Agent Architecture sebelum menyentuh agent/prompt.