One AI-built website wired to every channel your customers reach you on — with booking, CRM, journey automation, and an operations layer that works while you sleep.
Strategic audit, May 2026 — identified four features that take 12-18 months to ship at quality. They're built. Plus six more competitive moats, ranked here by depth and shipped status. Hover any row to see the inner workings.
Multi-signal orchestrator scans every 30 minutes through 10 rules — no-show risk, winback, review-request, demand-dip, waitlist-match, repeat-question escalation. Scores 0–100. Optional LLM rerank (Claude sonnet-4-6) for qualitative reordering. Auto-execute mode for 3 safe allowlisted rules.
Walks a new business through Stripe, Resend, Google Calendar, WhatsApp Business setup via a guided conversation. The browser-leg fills the forms — approval-gated, never autonomous. AES-256-GCM credential vault keeps OAuth tokens encrypted at rest.
Customer asks something your FAQ doesn't cover → F7 searches the web (Claude web_search_20260209 + web_fetch_20260309), synthesises findings, drafts a reply with citations. Auto-triggered on inbound WhatsApp / Telegram / Instagram. Owner approves before customer sees a word.
Returning customer messages in → F8-alt pulls prior memory (every emoji, every booking, every reply) AND does a fresh web lookup for context that may have changed (flight delays, weather, event cancellations). The owner's reply is informed by both.
Every action-emitting feature flows through a single gate. Hash-chained audit log, cooldown + rate-limit, per-session policies. Owner reviews and approves before anything customer-facing fires. 100% of D-features retrofitted.
Inbound webhook → classify message (question? returning customer?) → emit F7 + F8-alt actions to Approval Inbox. Closes the customer-flow loop end-to-end — no manual curl, no human-in-the-loop required to start the AI workflow.
Booking lifecycle (6), payment (3), review (3), customer lifecycle (10). Idempotent upsert on every fire. Memory-aware step kinds (D5) let journeys branch on customer history — same FAQ class 2+ times escalates to F7.
Owner Dashboard via WhatsApp, No-Show Predictor, VoC Digest, Review Drafter, Memory-Aware Journeys, Bilingual Greeting, Segments, Broadcast, Heatmap, Demand Forecaster, LTV Ranker, No-Show Recovery, FAQ Generator, Reschedule Wizard, Birthday Detector, Smart Waitlist, Cross-customer recs… and more.
WhatsApp Cloud API + Telegram Bots + Instagram DM (Meta Graph) + SMS (Twilio) + Voice handoff — all routed to one inbox with smart channel routing and bilingual auto-translate (Gemini Flash). Memory layer recognises customers across channels.
Single BrowserAdapter abstraction over Claude-Chrome, Playwright, Browserless, or self-host — swap providers is a 1-file change. PlaywrightAdapter implements navigate / runAgentTask / form-fill with 4-selector strategy. Approval-gated executor wired. Activation: one Railway command.
A salon owner runs the day out of WhatsApp, Instagram DMs, a paper booking diary, a reviews tab, and a card machine that doesn't talk to any of it. Conversations slip. No-shows are a surprise. Nobody remembers the customer who was here last month. The Conversational Commerce OS makes that messy front line into an actual system — without asking the owner to change how they already work.
Every way a customer can reach you, wired into a single dashboard view grouped by who they are — not which app they happened to use.
Meta Cloud API, connected directly — no reseller markup sitting between you and your customers.
Direct messages from your business profile land in the same inbox as everything else.
A free channel for the customers who prefer it — fully wired, no per-message cost.
Plain text messaging with carrier detection, for reach that doesn't depend on an app.
Transactional and conversational email, in the same thread history as the chat channels.
Inbound calls understood by AI and handed to WhatsApp with the context attached.
Unified inbox. Every channel grouped by customer identity. The owner replies once, and the system sends it back the way that customer prefers to be reached. Built-in auto-translate covers ten languages, cached so it stays fast:
Embeds on every TRD-built site. A submitted booking routes straight to the owner's WhatsApp and into the dashboard.
Bookings, calendar, reviews, channels, broadcasts, integrations, A/B tests, journeys, POS, costs and revenue — all in one place, all mobile-responsive.
Everything about one customer across every channel — booking history, the full message timeline, reviews they've left, and lifetime totals.
Most messaging tools fire one or two automated messages. The Conversational Commerce OS runs full multi-step journeys across the entire customer lifecycle — the capability that turns a chat app into a retention machine.
triggers, each one wired to a real event in your business — not a manual button someone has to remember to press.
A returning visitor is recognised across your website, your WhatsApp, and your booking form. The system pulls together what it already knows about that person and feeds it into two places: the way your site greets them, and the reply you send back in chat.
The result is continuity. The customer who booked three months ago isn't a stranger again — and you didn't have to remember a thing.
Visited the booking page twice last week
Asked about weekend availability in March
Left a 5★ review after the last visit
Fed into the site greeting and the WhatsApp reply — automatically.
Not chat features — operational ones. They run quietly in the background and surface only when there's something worth the owner's attention. Three groups, all shipped, all live in production.
Everything to do with answering a customer — drafting it, translating it, scheduling it, never losing it.
Run the business by texting it. Seven commands — today, confirm, no-show, revenue, customer, reminder, help — answered inside the chat the owner already lives in.
Drafts a tone-aware reply to every incoming review — gracious, neutral or apology, chosen automatically. The owner approves, edits, or discards. The writing is already done.
One paragraph each morning that synthesises the night's bookings, no-show risks, customer issues and drafts into a single read. Plus a ranked checklist of what to do today.
The same briefing, delivered to the owner's inbox every morning automatically. Idempotent, owner-opt-in, formatted for a phone read at the kitchen table.
Detects the customer's language from history (EN, HI, ES, FR, AR) and writes greetings side-by-side in English and their language — confirmation, reminder, thanks, apology.
Reads the top customer issue of the week and drafts a broadcast in the right tone — apologetic for severe, acknowledging for medium, upbeat for low — from a curated template bank.
What the system knows about the customer base — who's valuable, who's at risk, who's asking what, when the next quiet patch will hit.
Every booking gets a 0–100 risk score from 7 deterministic features. The owner sees exactly which customers to confirm — instead of finding out at appointment time.
Reads every customer message, sorts issues across 8 categories with severity scoring, and surfaces the single most important thing to fix this week.
Scans every customer's booking history and applies five segments — VIP, frequent, at-risk, new, dormant — by deterministic thresholds. Idempotent, so journeys can branch on it.
Aggregates every booking into a 7-day × 24-hour grid. The owner sees their actual peak and slowest hours, not a hunch.
Projects the next 7 days using the last 90 of day-of-week × hour history. Per-day expected bookings, peak hour, and a suggested staff number.
Ranks every customer by a composite value score: monetary, frequency, recency, completion ratio. Top 10% platinum, next 20% gold, next 30% silver, rest standard.
Scans every customer question, clusters similar ones together, and returns the top 20 with category and an answer template. The FAQ writes itself from the actual asks.
For every regular customer: median visit cadence, days overdue, churn risk (green / amber / red), predicted next visit and 90-day spend. Cohort totals roll up at the top.
What the system does on its own — winning back no-shows, rescheduling clashes, branching journeys based on memory, all without an owner touching a button.
Journeys that branch on what the system remembers — sentiment, visit count, segment, things the customer has said before. Five new step kinds layered onto the 22-trigger engine.
Two-stage winback. T1 fires 60 minutes after a no-show: "we missed you, reschedule?" T2 fires ~48 hours later with a discount code. The owner approves per-row before sending.
Detects reschedule intent in a customer message, then suggests the three best alternative slots — scored by demand baseline, day-of-week match and hour match.
The digests run, the scans run, the drafts get written, the predictions get scored — all on a schedule, in the background. The owner wakes up to a system that already did the night shift.
A strategic audit in early May 2026 identified four features that no SMB-chat platform currently delivers and that take 12–18 months to build at full quality. All four are now live in production with W1 + W2 in place.
Walks a new business through Stripe, Resend, Google Calendar and WhatsApp Business setup via a guided conversation. The browser-leg actually fills the forms — approval-gated, never autonomous. AES-256-GCM credential vault keeps OAuth tokens encrypted at rest.
Live · W1 Browser exec · W2 pending install
When a customer asks something your FAQ doesn't cover, F7 spawns a research action — searches the web (Claude web_search), synthesises findings, drafts a customer-facing reply with citations. Auto-triggered by inbound WhatsApp / Telegram / Instagram. Owner approves before the customer ever sees a word.
Live · W1+W2 Auto-trigger live
For returning customers, F8-alt pulls prior memory (every emoji, every booking, every reply) AND does a fresh web lookup for context that might have changed — flight delays, weather, event cancellations. The owner's reply is informed by both.
Live · W1+W2 Auto-trigger live
Multi-signal orchestrator. Every 30 minutes, scans a session snapshot through 10 rules (no-show risk, winback, review-request, demand-dip, waitlist-match, …), scores candidates 0–100, optional LLM rerank (Claude sonnet-4-6), emits top-3 to ★ Recommendations. Auto-execute mode for 3 safe rules (gated off by default).
Live · W1+W2+W3+W4 M6-9 capstone
Start free, move up only when the business does. Every tier is priced to be honest about what an owner actually gets for the money.
Honest status, no hand-waving. Here's what's actively being built into the Conversational Commerce OS.
BL-3-W2 Playwright adapter is shipped (Zone 123). One Railway command — npm i playwright — and the browser-leg goes from intent-only to actually filling forms.
Returning-customer auto-trigger is live. New-customer first-touch needs a phone_number_id → session map. Wired this session.
Telegram and WhatsApp inbound now fire the AT router. SMS handleInboundSms gets the same fire-and-forget hook next session.
Code path live and demoable in dummy mode. Real activation waits on Meta WABA approval queue. Embedded signup flow + phone verification both drafted.
Waiting on Stripe account activation. 5 SKUs (Hobby / Pro / Team / Business OS / Multi-site) ready to provision once approved.
F10-W4 is shipped (Zone 126). Set F10_AUTO_EXECUTE=true on Railway to flip — 3 allowlisted rules (waitlist match, onboarding nudge, stale cleanup) will skip approval for score ≥ 90.
F5 multi-vertical onboarding flows · F7 industry-specific source filters · F8-alt cross-customer pattern detection · F10 expansion from 10 → 30 rules.
POS integration moves from dummy data to live order and payment events, auto-linked to the matching booking.
Payment handling across currencies, for owners serving customers in more than one market.
The Conversational Commerce OS ships continuously. Most recent first.
Every customer message on WhatsApp Cloud, Telegram or Instagram now auto-classifies through the AT router (Zone 120) — F7 research and F8-alt memory lookups queue automatically in ★ Approvals. F10 grew to 10 rules with LLM rerank and auto-execute mode. The PlaywrightAdapter (Zone 123) is shipped and waiting on a one-command Railway install for full browser execution.
F10 — the audit's "single most defensible feature" and M6-9 quarter capstone — shipped W1. Multi-signal orchestrator scans every 30 minutes, scores rules, emits top-3 to Approval Inbox. All four GENUINELY UNIQUE moats now have working W1 in production within 30 hours of the audit incorporation.
The Approval Inbox protocol (Zone 103) became the gate every action-emitting feature flows through. F5 Smart Onboarding Bot, F7 Deep Research Agent, F8-alt Customer Memory all shipped W1 within the same 24-hour window.
Owner Dashboard via WhatsApp, No-Show Predictor, Voice-of-Customer Digest, Review Drafter and Memory-Aware Journeys — all live, all running on a schedule.
Two new verticals join Restaurant — with their own vocabulary, integration stubs and marketing pages.
The journey engine expanded from a handful of triggers to the full 22 across booking, payment, review and customer-lifecycle events.
Returning customers are now recognised across website, WhatsApp and booking form, with context fed into both the build and the reply.
The five-tier structure landed — from Free up to Business OS, which replaces three to five separate SaaS subscriptions.
The Conversational Commerce OS doesn't stand alone — it's wired into the rest of the TRD Network. Each part makes the others worth more.
A multi-agent system that turns a prompt into a finished, multi-page website across 22 industries — the site your Commerce OS runs on.
Explore TRD Platform →A GPU marketplace that supplies the inference power behind every build and every AI reply — at the economics only owned infrastructure allows.
Explore TRD Compute →Every TRD-built site is mirrored to Filecoin with an authorship certificate — so what you publish stays provably yours, permanently.
Explore TRD Storage →Start free, build your site, and connect the channels your customers already use.