# What we actually have — plain-language inventory (2026-08-19)

Written for James after the clinic-vertical veto. Purpose: you said you can't imagine what to do next because you don't know what exists. This is the full answer, in capability terms, not code terms. Read time ~4 minutes. Everything here is committed and real unless marked otherwise; nothing is clinic-specific unless marked 🏥.

---

## 1. The engine (the durable asset — vertical-agnostic)

**A software factory that mass-produces Thai-speaking AI business assistants.** One command mints a new branded app for a new business type. Four have been minted already: a wellness-spa template, 🏥 a clinic demo, a barbershop demo, and a Cameo-style talent-booking demo — each took minutes, not weeks.

Every minted app comes with, working today:

- **A Thai AI receptionist** that answers ONLY from that business's own data — its real prices, services, hours. It refuses discounts it wasn't given, refuses medical/expert advice, refuses questions outside its knowledge, and resists prompt-injection. This is not a system prompt hoping for the best — see §3.
- **Consequences with a human in the loop:** the AI can *request* a booking/order, never confirm one. A staff member confirms. Every action is audit-trailed.
- **Staff tools, phone-first:** admin screens, a conversations inbox where staff can take over any chat live from the AI mid-conversation.
- **An embeddable widget:** any business website can carry the chat with one script tag.
- **Known unit economics:** every AI reply is metered — ~฿0.4 per customer message. You can quote your own costs to the satang.
- **Multi-tenant core:** one deployment can serve many businesses with hard data isolation (bank-grade row-level security).

Modules that exist and compose into any mint: catalog/pricing, bookings, customer CRM, invoices/sales documents, tasks/workflows, chat guards, documents/files, marketplace orders. A billing/subscription system for charging *your* customers is fully designed for Thai payment reality (PromptPay/transfer first, direct-debit later, withholding-tax handled) — designed, not yet built.

## 2. The trust system (your confirmed thesis, made tangible — vertical-agnostic)

This is the part that makes the engine *different* rather than just another chatbot wrapper:

- **A written doctrine:** ~1 page, 15 testable promises the product makes («ตอบจากข้อมูลของร้านเท่านั้น» — answers only from the shop's own data, never invents discounts, never plays doctor, one intent = one consequence, every baht traceable). Drafted in my words, awaiting yours.
- **An automated trust harness:** one command replays 33 realistic Thai customer conversations against a live app and checks every reply against those promises. First live run: **138/141 checks held.** It caught the one known bug (double-booking on repeated confirmation — fix is queued and now safe to do) and even caught a subtle judgment call (may the bot say "you save ฿2,600" when that's derived arithmetic?).
- **A committed trust report** — evidence you can hand to a skeptic: here are the checks, here's what held, here's what we caught.

No competitor in the Thai SMB-AI space we researched shows anything like "we test our promises." This is a defensible story for ANY audience — customer, partner, or investor.

## 3. Go-to-market artifacts (built once, re-skinnable)

- **An interactive bilingual product tour** — polished offline HTML with real conversation replays. 🏥 Current content is clinic-flavored, but the tour *machinery* (template + build script + capture pipeline) rebuilds it for any vertical from new transcripts in about a session.
- **Sales kit:** pricing-sheet structure, one-person-one-hook outreach scripts, demo runbook with failure fallbacks, a 15-minute "ready to pitch" checklist. 🏥 The prospect list (Bangkok clinics) is the only truly discardable piece.
- **Market research, verified July–Aug 2026:** Thai SMB SaaS pricing landscape, competitor scan (Zaapi, Oho, FoxConnect, EMRPad…), Thai payment rails deep-dive (why cards fail here, why PromptPay/DD win), tax/WHT mechanics, the depa 200%-tax-deduction lever for buyers. Most of this transfers to any Thai SMB vertical.
- **A portfolio survey of your 10 other repos** — what they rehearse, what's reusable (e.g., proven Expo/SwiftUI mobile capability, the Luna quality-flywheel methodology that became the trust harness).

## 4. What does NOT exist (the honest gaps)

1. ~~**Nothing is deployed.**~~ **CLOSED same day (2026-08-19, spec §9.46/§9.48/§9.49):** the build was fixed and the whole fleet now runs 24/7 on a hardened Germany VPS with real HTTPS domains at `6326638.xyz` — tailnet-gated (only James's devices) until he chooses public. Deploys are one `git push`. See `91_QUICK_KNOWLEDGE_BASE_INFO_HUB.md`.
2. **No LINE integration** — web chat + widget only. For Thai SMBs, LINE is where customers actually are. (Needs your LINE dev account; task parked.)
3. **No payment collection built** — the design exists; no code. First customers would pay by PromptPay/transfer manually, which is fine.
4. **No public website** for the venture itself (the "would-an-a16z-VC-stay-interested" copy bar is set; nothing written).
5. **No mobile app shipped** — but the factory + your paid Apple/Google dev accounts + the Expo experience in your portfolio make the Cameo-style talent app buildable on order.
6. **One known product bug** (double-booking) — reproducible on demand now, fix queued.
7. **No customers, no revenue, no external users.** Every asset above is inward-facing until you point it at someone.

## 5. What each revenue shape would draw from this

**These three are NOT exhaustive** — they are simply the shapes James happened to type off the top of his head (2026-08-19); he expects proper thinking to produce more and better ones. The table shows what the inventory contributes to each *known-so-far* shape; a new shape gets its own row when it appears.

| If first revenue turns out to be… | What we already have for it | What's missing |
|---|---|---|
| **Consulting / paid expertise** ("help me adopt AI safely") | The doctrine + harness + trust report = instant credibility artifact; the portfolio shows range; the factory becomes your delivery accelerator | A one-page positioning of YOU (not the product); someone to say it to |
| **Subscription** (a business pays monthly) | The whole engine §1; sales kit structure; pricing research; billing design | A chosen vertical/first customer; deployment (gap 1); eventually LINE (gap 2) |
| **Fundraise** | The factory story + trust thesis + demo fleet + unit economics; the side-by-side "same AI, without our rails" demo is one session from existing | Deck, website (gap 4), and the traction story a raise usually wants |

Deliberately no recommendation among them — the shape stays intentionally vague (could be ฿100 or ฿1M; consulting, subscription, a raise, or something not yet imagined), and nothing we'd build next for one forecloses the others. Notably: **fixing deployment (gap 1) serves every shape listed so far.**

---

*Repo: `/Users/james/3_nodejs/ai-new-business` · task menu: `00_TOC_TRACKER_REMAINING_WORK_OPUS_FABLE.md` · resume prompt after compaction: `01_FOR_JAMES_TO_COPY_PASTE_THIS_PROMPT_18AUGUST2026.md` · this file: written 2026-08-19 during the post-clinic-veto re-alignment.*