Our Meta route explained, how Sellebrate is probably wired, and how Spanish Homes Online can go live before the approval lands.
10 October 2026 For Stevan and Tom. Private link, not indexed.
What we are building has two official names stacked together. We are becoming a Meta Tech Provider: our own Meta app, "Agency AI" (1404769571184582), asking for advanced access to the two WhatsApp permissions, messaging and management. And we are doing it through Twilio's Tech Provider Program, where Twilio stays the infrastructure and billing layer. Once Meta approves, Twilio links our app to their Partner Solution, and from then on an agency onboards itself on our page: one Facebook login in Meta's popup, and Twilio registers the number behind the scenes.
The submission went in on 1 October at 20:15 UTC with both permission checks green. Here is the whole pipeline:
Bot live on the devbox sandbox, 287 tests, signup page coded and previewed.
Both screencasts, usage descriptions, API test calls all accepted.
"Most submissions are reviewed within 20 days." Decisions often arrive inside a week.
About 5 business days. Business verification of CurrencyTransfer is already done.
"Connect your Meta app to the Twilio Partner Solution." 1 to 2 business days, then we accept it.
Embedded Signup config, stable https host for /onboard, first agency connects itself.
Realistic distance from today to the first agency onboarding through our own page: two to three and a half weeks, dominated by the two Meta queues. Nothing about our submission is abnormal or behind schedule.
This is the piece that reframes everything else in this document. Meta's permission model has two tiers, and the difference is not "can you use the API".
Standard Access is automatic for every business app. It lets an app message from any WhatsApp Business Account owned by the same business that owns the app. Meta's own wording: "If you are using the API for yourself as a Direct Developer, you do not need Advanced access or app review." Thousands of businesses run production WhatsApp this way, with no review, ever.
Advanced Access, the thing we are queuing for, unlocks exactly one capability: our app touching WhatsApp accounts owned by other businesses. Client accounts return error 200 without it. That is the whole Tech Provider construct, and it is what a white-label signup page onboarding agency after agency legally requires.
The review was never gating "using WhatsApp's API". It gates serving other businesses' WhatsApp accounts from our app. One agency, set up by hand, needs no review from anyone. Twenty agencies, self-serve under our brand, need the review.
One constraint shapes every option below. We decided early, for good reasons, that the agency is the sender: messages come from "Spanish Homes Online", never from CurrencyTransfer. Name mismatch, no opt-in and block risk all sit on the wrong side otherwise. So whatever route we take, the WhatsApp account must be the agency's, which from our app's point of view makes it a client account. Hold that thought.
Sellebrate (Sellebrate, Sociedad Limitada; founder Gerard Barrull) is the closest competitor we have found: an AI agent for Spanish estate agencies, positioned as "an AI agent, not a new CRM". What their public pages actually establish:
Nothing public names a provider, which itself narrows the field. In October 2026 there are only two realistic shapes, and both are structurally what we are building:
Route one: their own Tech Provider app, exactly our play, either over a BSP or straight onto Meta's Cloud API with nobody in the middle. The "Meta's official API" phrasing next to an OpenAI stack suggests a lean direct integration is plausible; it is also the cheapest way to run this at their price point.
Route two: riding a BSP's partner program, where a provider such as 360dialog hosts the signup under its own Meta app and Sellebrate never faced App Review at all. Lower lift for a small Spanish SL, paid for in margin and lock-in, and an arrangement the BSPs themselves are now migrating away from by pushing partners to register as Tech Providers.
Their "under a week, you set up nothing" is almost certainly the same concierge move we already planned for Adam: a screen-share call where the agency does one Facebook login. Either way, the conclusion for us is identical. What we are queued for is something they already have. The queue proves the barrier is crossable; it is a toll, not a wall.
If we want certainty, a demo call (Tom posing as a prospect works) gets it in ten minutes:
Their private-seller module cold-messages numbers scraped from portals. On the official API that is opt-in and quality-rating exposure, plus a portal terms-of-service problem. Our backlist rules (own-portal leads only, 6 and 12 month windows) are the version of that feature that survives scrutiny. Worth remembering if an agency ever compares us feature-for-feature.
Honest answer: they are ahead on breadth and go-to-market; we are ahead on depth where it counts for our niche. And our depth is measured, while their claims are marketing copy.
| Axis | Sellebrate | Callum AI |
|---|---|---|
| Status | Live with Spanish agencies, onboarding in under a week | Built and tested; platform gated on Meta review, pilot path open now |
| Surface | Seven modules across WhatsApp, Instagram, web, matching, prospecting | One job: re-engage, answer, qualify, hand off a noted lead |
| Integrations | Portals plus six CRMs | None yet; lead-email ingest is next on the plan |
| Conversation guardrails | None evidenced publicly | Enforced in code: no invented claims, no fake promises, no menus, coverage honesty, question budget |
| Measured quality | Unpublished | Graded sims, 12 real personas; truthfulness 3.50 to 4.58 across Wave 2; 287 tests |
| Market DNA | Spanish domestic sellers and buyers, mandate-winning first | International buyers in English: airports, viewing trips, buying process, SMS fallback |
| Commercial model | Software fee, "a few hundred a month" | FX economics on the purchase itself; fee is the fallback, not the business |
Three things deserve emphasis. First, the generic failure modes of an OpenAI wrapper, inventing stock, over-promising callbacks, interrogation-style questioning, are precisely what we spent five weeks instrumenting out; "qualifies in your brand voice" is the claim every wrapper makes. Second, their whole machine is tuned for Spanish domestic agencies; Spanish Homes Online's leads are British buyers, and nothing in Sellebrate's offer touches that conversation's DNA. Third, their pricing line quietly validates our own commercial analysis: the fee band we modelled sits exactly where they price.
Worse on breadth and maturity. Better on the core conversation and the niche. Same plumbing underneath, where they are simply through a gate we are queued at.
Their existence validates the category and puts a clock on us: the two gap-closers already on our plan, the backlist importer and lead-email ingest, are buildable during the review wait.
Stevan's question this week: can we just use Meta's Cloud API directly instead of waiting? Yes, the Cloud API works today with no review, but only where the app and the WhatsApp account belong to the same business (section 2). Since the sender must be the agency, that constraint leaves three real paths to a live pilot:
| Path | Meta review | New code | SHO live | Reading |
|---|---|---|---|---|
| Wait for approval, then /onboard | Yes, in progress | None | 2 to 3.5 weeks | The platform play. Right for agencies two onwards. |
| Twilio sender for SHO now | None | None | About a week | Twilio's own signup creates SHO's account; our existing stack works unchanged. Recommended. |
| Direct Cloud API, app in SHO's portfolio | None | 1 to 2 days plus ops | Next week | Saves Twilio's roughly 0.4p per message. Buys nothing else; forks the stack. |
It was already our documented plan before the Tech Provider work started. Twilio's console runs Meta's signup under Twilio's own approved app: Adam does one Facebook login on a 15-minute call, SHO's WhatsApp account is created under SHO's own portfolio, and the sender lands in the Twilio stack our bot already speaks. Same webhooks, same send path, every test still meaningful. An unverified business can send 250 business-initiated conversations a day, which comfortably covers the pilot: a 50 to 100 message backlist run plus tens of daily leads, and replies inside the service window do not count against it.
The direct Cloud API alternative means a new adapter (different webhook format and signatures, Graph send calls, template management), a per-tenant Meta app created inside SHO's portfolio, and a payment method on SHO's account for template sends. All real work, all maintenance, to save a rounding error at pilot volume, in the exact week we want stability.
A portal enquiry does not open a WhatsApp messaging window. The free-form 24-hour window opens only when the buyer messages us on WhatsApp. So the first touch to a portal lead is always an approved template, on every route, on anyone's infrastructure, Sellebrate's included. That is Meta's rule, not Twilio's. Everything after the buyer's first reply is free-form, which is where the bot lives.
Direct Cloud API becomes genuinely interesting then: a Tech Provider with advanced access can run client accounts straight on Meta's infrastructure with no Twilio in the middle, on the very permissions we are already waiting for. That is a margin decision for later, not a way around the queue, and nothing chosen this week closes it off. SMS fallback keeps Twilio in the stack regardless.
Meta: App Review for solution providers, WhatsApp permissions, Cloud API get started. Twilio: Tech Provider Program and integration guide. 360dialog: registering as a Meta Tech Provider. Sellebrate: WhatsApp CRM page, home, their 2026 guide. Verification limits: business verification and the 250-a-day unverified cap.
Caveats: every Sellebrate statement is from their marketing pages as read on 10 October 2026 and is unverified; their pricing is a line of copy, not a rate card. The "development mode accepts but does not deliver" behaviour is community-corroborated rather than cleanly documented by Meta. Internal numbers (tests, sim scores, latency, costs) are from the project's own measured runs and analysis pack.