WhatsApp & Instagram AI Assistant

What it does

An AI assistant that plugs into a business's WhatsApp and Instagram accounts, talks to customers in natural language and performs real actions. It doesn't just answer FAQs: it creates orders, books appointments, checks shipping status, saves leads and hands the conversation to a human when it should. Everything it does lands in the business's CRM panel in real time.

How it works

• Tool-using agent: an agent → tools → agent loop built on LangGraph. Ten tools (orders, appointments, stock and price, leads, human handoff…) do their work directly in PostgreSQL.
• Per-business tool set: each business record stores which tools are enabled and the agent only sees that subset. One engine, an appointment storefront for a clinic and an order storefront for e-commerce.
• Confirm before writing: every tool that writes to the database summarises first, gets the customer's confirmation, then runs.
• Channel-agnostic core: the WhatsApp and Instagram routers only parse the Meta payload; conversation handling, human takeover and the agent call live in one shared core.
• Human handoff: on complaints or refunds ai_paused is raised, the AI goes quiet and staff continue from the panel.
• Live panel: the CRM receives events over WebSocket, so an order appears the moment it is created.

Tech

Python · FastAPI (async SQLAlchemy) · LangGraph · PostgreSQL · React · TypeScript · Tailwind · WebSocket · Docker Compose. The model provider is switched in .env (Google Gemini / Anthropic Claude). Real integration runs through the WhatsApp Cloud API webhook.

Status

At demo stage and multi-tenant: every prospect can get their own demo account with its own business profile, catalogue and channel entitlement.

A detailed write-up of how it was built: An AI that answers on WhatsApp: from conversation to a real order.