Radical transparency
Where the AI is.
And where it isn't.
Every AI company implies their software is a thinking machine watching over your business. Most of it is ordinary code. We think you deserve the honest map — so here is exactly how every part of Loggia works, feature by feature. This page is checkable against the product.
The five mechanisms
| Class | What it means |
|---|---|
| AI-in-loop | Claude runs automatically inside the flow, no human trigger. Used sparingly, metered per organization. |
| AI-on-demand | Claude acts when you click, and its output is a draft you approve. It never sends, spends or publishes on its own. |
| Deterministic | Plain code — timers, webhooks, database writes. No AI anywhere in it. Fast, auditable, boring on purpose. |
| Inert-until-keys | Fully built, does nothing until the relevant integration (Stripe, Twilio, email domain) is configured. |
| Simulated | Demonstrates the full flow without performing the real external action. Always labeled in-product. |
Feature by feature
| Feature | Mechanism | What actually happens |
|---|---|---|
| Customer chat widget | AI-in-loop | Claude answers automatically from your uploaded documents. One click hands the thread to a human and the AI stops. |
| Ad campaigns | AI-on-demand | Claude drafts audiences, headlines and budgets. Nothing runs until you approve, spend caps are armed first, and there's a kill switch. |
| Follow-up emails | AI-on-demand | Claude writes the draft into your composer. You press send. |
| Booking follow-ups | AI-on-demand | When someone books, Claude drafts a personalized follow-up as a task for your approval. The booking itself involves no AI. |
| Booking → calendar | Deterministic | A booking writes the contact, appointment and activity in one database transaction. It appears "instantly" because nothing is watching — it's a direct write. |
| Appointment reminders | Deterministic | An hourly job reminds everyone with an appointment in the next 24h, exactly once, by email or SMS. |
| Drip sequences | Deterministic | The delays and templates you configured, sent by a timer. Unsubscribes are honored before every send. |
| Lead attribution | Deterministic | Arithmetic over your own contact sources, including UTM tags captured automatically. No AI, no estimates. |
| AI Visibility scan | Deterministic | We fetch your public pages and check schema, crawler access, sitemap and content — every finding is verifiable in your own page source. |
| AI Visibility fixes | AI-on-demand | Claude writes the schema and FAQ markup from your real business record. You review and paste. Nobody can buy a ChatGPT ranking — we make you legible, the mention stays the model's choice. |
| Autopilot: speed-to-lead | AI-in-loop | This one sends without you. Off by default. Once you switch it on and write the opening message yourself, a new lead gets that message within seconds, personalized by Claude but never beyond what you wrote. Quiet hours, one message per person per day, unsubscribes honored, every send in your audit log, one switch pauses it all. |
| Autopilot: AI appointment setter | AI-in-loop | Also sends without you, when enabled. Claude replies to texts using your uploaded documents and steers toward a booking — and hands the conversation to you the moment anyone asks about price, complains, asks for a person, or it isn't confident. Handoffs become tasks so nothing is dropped. |
| Two-way SMS | Inert-until-keys | Real Twilio sends and signature-verified inbound webhooks once configured. Until then, the tab explains what's missing. |
| Billing | Inert-until-keys | Stripe-hosted checkout — card numbers never touch our servers. Plan status is written only by Stripe's signed webhook. |
| Linked publishing | Simulated | Uploads, transcodes, captions and your bio page are real today. The final publish to TikTok/YouTube/Instagram/X/Threads/Bluesky runs in a labeled demo mode while platform API reviews complete. |
The rule behind all of it: the AI proposes, a human commits — with exactly one deliberate
exception, named above. Autopilot can text a new lead without you, because a reply that waits for
approval isn't a fast reply, and speed is the entire point. It is off until you turn it on, it can
only send the message you wrote, and one switch stops it. Nothing else in Loggia spends money,
contacts a customer, or publishes content without explicit human approval — and every AI action is metered and logged per organization.
Why publish this?
Because "AI-powered" has become unfalsifiable marketing, and small business owners can't audit code. A mechanism map is something you can check: click the features, watch what they do, and hold us to it. If this page ever disagrees with the product, one of them gets fixed the same week.