ELAPS

Build • Innovate • Evolve

ELAPS
ELAPS — Build • Innovate • Evolve
Back to Products
Live

ServeBot

WhatsApp Ordering Assistant for Restaurants

ServeBot is a WhatsApp ordering assistant for restaurants. Guests order by text or button taps — Level-1 routing handles menu, cart, checkout, and order status without an LLM. Anything open-ended goes to a shared LangGraph agent with scoped tools. Owners get a Next.js dashboard with live orders and an AI assistant for sales & ops.

Request a Demo Download docs · soon

Your guests already message on WhatsApp. ServeBot takes the order, updates the cart, sends payment links, and answers menu questions — while your kitchen dashboard stays in sync and your admin AI answers “how many pending orders right now?”

Why teams choose ServeBot

  • WhatsApp-native ordering — interactive lists & buttons, no guest app install
  • Level-1 rule router — menu, cart, checkout, status with zero LLM cost when possible
  • Level-2 LangGraph agent — live_chat tools for guests, dashboard tools for owners
  • Admin dashboard — orders, menu CRUD, customers, announcements, AI chat
  • Postgres checkpointer — multi-turn memory across WhatsApp and dashboard
  • Meta Cloud API + FastAPI + Next.js — production three-service architecture

Agentic Workflow Architecture

Level-1 rule router for menu, cart, checkout & status — Level-2 shared LangGraph agent with chat tools for guests and dashboard tools for owners. Checkpointer keeps every thread.

Live path · Menu QA

LEVEL 1 ROUTER · rule-based / button payload · no LLMLEVEL 2 · shared LLM agent · tool access scoped by entry flowsearch_menu · get_menu · edit_cart · checkout_link · order_status · RAGsummary · sales · inventory · availability · top sellersyesyesno · doneSTARTIncoming messageWhatsApp text / button tapintent_routerkeyword / button payload · no LLMmenu_qaDB · FTS querycart_editdirect DB writecheckoutpayment linkorder_statusDB lookuplive_chatmenu · cart · order Q&Alive_dashboardsales & ops · NEWagentLLM reads state · picks tool or donetool call?chat-flow toolsmenu · cart · checkout · RAGdashboard_toolssales · inventory · NEWrespondextracts / formats replysend_replyWhatsApp Cloud API callENDcheckpointer persists conversation state — waits for next message
LLM / routing
Deterministic / DB
Payment
Dashboard tools

Key Capabilities

01

Two-level agentic routing

intent_router matches keywords and button payloads first. Deterministic paths hit menu_qa, cart_edit, checkout, and order_status. Unmatched or “Ask Anything” falls through to the shared LLM agent — fast when rules apply, smart when they don’t.

02

Guest live_chat tools

search_menu, get_menu, edit_cart, checkout_link, get_order_status, and knowledge-base RAG (pgvector) — the agent loops until the reply is ready, then send_reply hits WhatsApp Cloud API.

03

Owner live_dashboard AI

Dashboard Chat uses the same agent with scoped tools: summary, sales reports, inventory, menu availability toggles, and top-selling items — without exposing guest cart tools.

04

Restaurant admin UI

Next.js dashboard for today’s stats, orders, menu CRUD, customers, WhatsApp announcements, settings, and the AI assistant — JWT-secured against the FastAPI API.

05

Durable conversation memory

LangGraph AsyncPostgresSaver checkpoints per thread (whatsapp:<phone> / dashboard:admin) so multi-turn carts and admin chats survive restarts.

06

Production Meta WhatsApp stack

Signature-verified webhooks, wamid dedupe, interactive list/button payloads, and agent-driven replies via Meta Graph API — plus optional Langfuse tracing.

Built With

LangGraphFastAPINext.jsPostgreSQLOpenAIWhatsApp Cloud APIpgvectorSQLAlchemyLangfuse

At a Glance

WhatsApp ordersLangGraph agentAdmin AIPayments

Ready to try ServeBot?

See how ServeBot fits your team. Request a demo — or grab the product brief when it's ready.

Request a Demo Download docs · soon
Chat with ELAPS · or play ELAPS Rush