AI support that knows the brand it is working for.
A multi-tenant customer support platform that adapts to each brand's unique voice and knowledge. The AI learns from the brand's own website, delivering tailored responses across web, WhatsApp, and voice channels.
AI support that learns each brand's voice from its own website.
Vereby is a multi-tenant customer support platform. Every brand gets its own knowledge base, scraped from the brand's actual site on day one. Voice, fallback phrases, and greetings are inferred from the first 3,000 characters of that scrape, so a new brand opens with a tone that already matches.
The result is a customer support agent that reads like the brand's own team member: knows the products, knows the policy, knows the voice, and answers the same way on web chat, WhatsApp, and voice.
Three use cases, one brain.
Vereby is a multi-tenant customer support platform. Three live use cases, one AI brain that adapts to each brand.
Ecommerce. Salons. Plumbers. The integration layer changes (Shopify, Calendar, Calendar with emergency dispatch). The voice, the knowledge, the per-brand training, the channels: all configured per brand. The widget looks the same. The voice is different.
Ecommerce. Real Shopify conversation. Order tracking, inventory lookup, returns, all from the live store.
For the store.
Order tracking, live inventory, returns, and policy questions. Pulled from Shopify in real time. When the customer asks "where is my order #1042", the model calls order_lookup and gets back "Shipped via Royal Mail. ETA tomorrow by 3pm. Tracking RM123456789." Real data, not vibes.
Returns and policy questions route through the same brain, with the brand's own policy text in context. WhatsApp, web, and voice on the same model.
For the diary.
Books cuts, colours, and treatments straight into the calendar. Sends reminders. Reschedules without the back-and-forth. The model checks live availability across stylists, knows the service prices and treatment times, and books a slot that fits.
WhatsApp confirmations and reminders come from the same conversation, so a customer can change a booking without picking up the phone.
Salons. Real booking conversation. Calendar integration. Service prices and treatment times baked in.
Trades. Real emergency dispatch. Triage, route, book, all from a single WhatsApp message.
For the van.
Takes the call, triages emergencies, books the job, routes the urgent ones to whoever is on. Twenty-four seven, no missed leads. The model classifies the call, picks the calendar, finds the slot, and confirms the booking.
WhatsApp first. Voice fallback. Web form last. The customer uses whichever channel is closest, and the brand gets the same answer every time.
Three services, one Supabase project.
| Customer backend | FastAPI on Railway, port 8000. Serves the public site, the widget, the MCP endpoint, and the built admin SPA. The single public entry point. |
|---|---|
| Admin backend | FastAPI, port 8001. The dashboard API: brands, channels, knowledge, conversations. The operator's surface for configuring each brand's brain. |
| Admin frontend | React 18 with Vite, dev port 5174. Built into the customer backend's public/admin/ for production. One deployment, two audiences. |
| Database | One Supabase project. Brands, channels, knowledge, conversations, voice profiles. Per-brand row-level isolation, fed by Zoho-style webhook architecture in spirit if not in detail. |
| Scraping | httpx for static pages, BeautifulSoup for text extraction, Playwright headless shell for SPAs, sitemap.xml auto-discovery when internal links are absent, LLM batch extraction (4 pages per batch, 2 LLM calls per batch for FAQs and chunks). 5-strategy JSON parser handles model output reliably across formats. CloakBrowser is a drop-in for sites that block stock Playwright. |
| Voice auto-generation | When a brand is scraped for the first time, the AI samples the first 3,000 characters of scraped content and infers tone, greeting, fallback, and goodbye. New brands get a proper voice on day one. |
One platform, one brain, one voice per brand. The AI learns the brand, the orchestration does the work, and the operator keeps the keys.
Free, Basic, Starter, Growth.
A real free tier because the cheapest way to prove the product works is to let a small brand run it for thirty days. Pricing scales with the work the AI is doing, not the seat count, because the work is what costs money, not the login.
Four tiers. Thirty-day trial. No card required to start.
The same build approach for any complex AI system.
Hermes powered workflow
Vereby's multi-tenant architecture is orchestrated through Hermes. The same agent system handles knowledge ingestion, channel routing, per-brand voice configuration, and the tool calls behind every answer. When a model improves or a brand's needs change, the swap is a config change, not a rewrite.
Right-sized model for the task
I route economical LLMs to the high-volume work (initial triage, information retrieval), and reserve frontier models for the moments where nuanced understanding and accurate tool use is the actual job.
Rigorous quality assurance
The build process includes automated quality checks and frontier-model audits covering backend robustness, data isolation, and consistent functionality across every integrated platform, from web widgets to WhatsApp.