LBD Business Comm
The customer communication platform that brings together an omnichannel inbox, a web widget and an AI that answers for every type of user — with a seamless handover to a human when it matters.
Multi-tenant · multi-audience · realtime · already running on our own projects
Fig. 01 — Web widget · Shadow DOM
Shadow-DOM web widget, instant load
Web, WhatsApp, Messenger, Instagram, Email
Segregated AI knowledge and answers per user type
Data and infrastructure entirely in the European Union
Inside the product
The same screens our teams use, redesigned: the agent panel, the end-user widget and the AI that acts as the first line.
You’re welcome! If you have any…
Booking number: we’re thrilled…
Perfect! We’ll pick you up at 15:15…
Hello, we are interested in your…
Perfect, thanks for sorting that…
Only have the 4th at Catania…
Happy to help! We offer several Mount Etna tours:
Half-Day & Full-Day Jeep Tours — lava flows, caves and volcanic landscapes off-road.
Etna Summit Trekking — 4×4 up to 2,800 m, then trek to ~3,300 m.
What interests you most? I can detail price, duration and what's included.
Queues by view (all, unassigned, mine, by team), contact profile, bookings, conversation history, internal notes, translation and AI reply generation.
Web messenger in Shadow DOM
A home with help center, contact cards and search; a conversation with AI bubbles, in-chat data collection and continue-on-WhatsApp. Customizable for appearance, persona and audience.
Answers, collects data, hands off to a human
The AI answers with the right knowledge for each user type, collects the data it needs and, when required, alerts agents by handing the conversation over without losing context.
Fig. 03 — Widget · WhatsApp handoff
One conversation, every channel
Your customers write wherever they like: everything lands in one inbox, with the same AI and the same history.
Everything you need to talk to your customers
A multi-tenant foundation built to scale: every capability is designed so later phases are additive, with no rewrites.
Per-audience AI with RAG
Answers generated on each user type’s segregated knowledge, with semantic retrieval and re-ranking.
- Claude: Haiku to classify, Sonnet to answer, Opus for the hard cases
- Voyage AI embeddings + rerank, vector search on pgvector
- A “WHEN → THEN” behavior flow evaluated before the LLM
Multi-tenant & multi-audience
Two first-class dimensions in the data model: workspace for the tenant and user type (e.g. customers / partners) for the audience.
- Tenant isolation with Postgres RLS as a safety net
- A different help center, knowledge and answers per audience
- A “partner” document never ends up in an answer to a “customer”
Realtime agent inbox
A dashboard where agents see, claim and answer conversations in real time.
- All / unassigned / mine / per-teammate / per-team views with counts
- Presence, typing indicator and read receipts
- Internal notes, macros, translation and AI reply generation
Omnichannel
The same thread, whatever the starting channel, with shared history and AI.
- Web widget, WhatsApp, Messenger, Instagram
- Inbound and outbound email with threading via AWS SES
- Continue a web chat on WhatsApp with one tap
Routing & assignment
Conversations reach the right person automatically, following configurable rules.
- AI-first flow with a fallback to “unassigned” or to a team
- WHEN audience / channel / contact attribute THEN team rules
- Round-robin or balanced auto-assignment within the team
In-chat data collection
Email, name and phone collected inside the conversation, with no external forms.
- The AI emits a form rendered in-chat by the widget
- On WhatsApp the customer replies in free text, the AI parses it
- Configurable timing: on start, before handover or both
Configurable widget
A micro-bundle isolated in Shadow DOM that never interferes with your site.
- Vite + Preact, under 50 KB gzipped
- Appearance, home-card and persona configurator with live preview
- Internationalization and RTL support
Per-audience help center
Self-service articles and search right inside the widget, different for each user type.
- Content and knowledge segregated per audience
- Search built into the messenger home
- The same base that feeds the AI’s answers
Identity & security
The audience of whoever writes is signed by your backend and cannot be spoofed by the client.
- End-user identity via HMAC or tenant-signed JWT, anonymous lead as fallback
- Agents authenticated with Keycloak (OIDC), roles owner / admin / agent
- Postgres RLS defending tenant isolation
Mobile SDK
Bring the same chat experience inside your native apps.
- Flutter SDK with realtime over Socket.IO
- Push notifications (Firebase) and local notifications
- Same signed identity and same per-audience AI
Architecture & stack
Built on solid, mostly open-source technologies, on European infrastructure.
A roadmap of additive phases
The data model is designed so each phase adds to the previous one, with no rewrites.
Multi-tenant, web widget, realtime inbox, per-audience AI RAG with handover, help center and mobile SDK.
WhatsApp, Messenger, Instagram and email in a single inbox.
Ticket management and service levels with deadlines.
Configurable automations and work flows.
Proactive messages and outbound campaigns.
Analytics, conversation quality and advanced CSAT.
Copilot-style AI with tool-use for operational actions.
Self-service, plans and platform billing.
Want to see it on your own channels?
We’ll show you LBD Business Comm in action and figure out together how to fit it into your customer communication flows.