leadbox.site
AI social inbox for Instagram, WhatsApp and Messenger — one shared workspace where an AI agent trained on the business's own website, catalogue and FAQs answers customers around the clock, and the team can step in at any moment.
Category
SaaS / AI Customer Messaging
Client
LEADBOX
Start Date
2026
platform
Vue 3, Vite, Python FastAPI, PostgreSQL, Firebase Auth, Meta Graph API, LLM agent, Caddy, Docker
Project Description
project description
Leadbox is an AI-powered social inbox for businesses that sell and support customers through Instagram, WhatsApp and Messenger. Every message from every channel lands in one inbox, where an AI agent answers product, stock, shipping and order questions in the brand's own voice — and hands the conversation to a human whenever it should.
THE STORY
Small and growing businesses now do most of their selling inside DMs. Customers ask the same questions again and again — is this available, do you ship, can I change my order — across three different apps, at every hour of the day. Messages get missed, replies come late, and sales quietly go to whoever answered first.
Leadbox was built so that no message goes unanswered. The business connects its Instagram, Facebook and WhatsApp channels in a few clicks, adds its website, product catalogue, policies and FAQs as knowledge sources, and switches the AI agent on. From that moment every conversation is answered instantly, grounded in the business's real information rather than generic AI guesses.
The team stays in control: they see every conversation live in a single inbox, can take over any chat, add notes and approve replies before they are sent — while the agent keeps working on everything else.
OUR APPROACH
The product was engineered around three things: one inbox for every channel, AI answers that are accurate and on-brand, and strict data isolation between businesses.
• Unified inbox: Instagram, WhatsApp and Messenger conversations arrive in one real-time workspace, with live updates pushed over WebSockets and filters for unread and per-channel views.
• AI agent grounded in the business: replies are generated from the business's own website, product catalogue, shipping and returns policies and FAQs, so answers stay accurate and in the brand's voice.
• Always on, with a fallback: customer replies go to an OpenAI-compatible LLM, with a second provider configured as automatic fallback so the inbox keeps answering even when the primary model is down.
• Human handoff: the agent escalates when it should, and any team member can take over a conversation, leave internal notes or approve replies before they go out.
• Bookings and requests: the agent can handle bookings and cancellation requests directly inside the chat and log them for the team.
• Multi-tenant by design: every registered business gets its own isolated account, resolved from the owner's verified Firebase identity, so no one ever sees another business's chats, settings or connected pages.
• Meta integration: Facebook pages and Instagram accounts are connected through Meta's official APIs and webhooks, and land automatically in the owner's own account.
• Landing page: a fast, hand-built HTML/CSS landing with an animated live inbox demo that cycles through all three channels, plus sign-up and log-in flows for the app.
• Delivery: a Vue 3 + Vite frontend, a Python FastAPI backend and PostgreSQL, routed through Caddy and deployed as a Docker stack on a VPS with Dokploy, auto-deploying on every push.