Prepared for Michael Smorenburg · What It Takes + Running Costs
Building the software is the fast part, and it is mostly the part we do. Turning it into a running emergency service takes more than a build, and it is honest to say so up front.
This is the full picture: the third-party services with monthly subscriptions, the accounts only a person can open, the partner deals with the towns and responders, the legal and privacy work, and the management it takes from your side while Lumanaire builds. With indicative running costs, a lean pilot versus a sovereign production system.
Prompting an AI builds a working app and a convincing demo quickly. It does not, on its own, create a running emergency service. The difference is everything around the code, and for something people's lives touch, that is exactly where the time and money belong.
The code is weeks of work and the cheap part. The accounts, subscriptions, partner deals, compliance and vigorous testing are the long pole and the real cost.
Lumanaire and the AI carry the build. Some things, by their nature, only a person can do, and some things are deals that move slower than any code.
These are third-party running costs only. They exclude Lumanaire's build fee, which is quoted separately. Two scope levels drive the bill: a lean cloud pilot to prove the loop, and a sovereign production system at full scale.
| Line item | Tier | Lean pilot | Sovereign |
|---|---|---|---|
| Fixed monthly subscriptions | |||
| Hosting Vercel | Core | $20 | $30 |
| Database Supabase / SA Postgres | Core | $25 | $300 |
| Auth + MFA Clerk / Supabase Auth | Core | $25 | $80 |
| Maps + tiles MapTiler | Core | $25 | $50 |
| Monitoring + uptime Sentry + uptime check | Core | $26 | $120 |
| Sovereign compute SA cloud region, redundant | Sovereign | – | $500 |
| Self-hosted STT (GPU) Whisper / MMS, fine-tuned | Sovereign | – | $800 |
| Usage, scales with call volume (monthly estimate) | |||
| AI engine Anthropic (Claude) | Core | $50 | $350 |
| Speech-to-text Google Chirp 3 / ElevenLabs | Core | $40 | $150 |
| Voice telephony Twilio / Africa's Talking | Core | $40 | $300 |
| SMS SMS gateway | Core | $20 | $200 |
| WhatsApp Meta BSP, 360dialog / Clickatell | Core | $40 | $300 |
| Dedicated shortcode Carrier, monthly | Sovereign | – | $500 |
| Monthly total | $311 | $3,680 | |
| One-off setup (not monthly) | |||
| Domain Have lumanaire.co.za | Core | $0 | $0 |
| Meta verification WhatsApp prerequisite | Core | $0 | $0 |
| Integration setup Provider wiring + config | Core | $150 | $150 |
| Shortcode setup Carrier onboarding | Sovereign | – | ~$3,500 |
| App store accounts Apple + Google, responder app | Fast-follow | – | $125 |
| STT fine-tuning GPU compute, local audio | Sovereign | – | ~$2,000 |
| Legal + POPIA External law firm | Sovereign | – | ~$6,000 |
| One-off total | ~$150 | ~$11,775 | |
Indicative estimates in USD to validate with each provider, not quotes. Local providers (Africa's Talking, Clickatell) bill in ZAR. Usage lines assume low pilot call volume and rise with real traffic. Sovereign one-off ranges: shortcode setup $2,000 to $5,000, fine-tuning $1,000 to $3,000, legal and POPIA $3,000 to $10,000 (external counsel). The full editable line-by-line model is in the accompanying spreadsheet.
Lumanaire wires every one of these into the app the moment it exists. Lumanaire cannot bring them into existence. Each needs a person, and often a card, a registered business, and identity checks.
Even with the AI doing the building, the human load is real and continuous. This is the part that is easiest to underestimate.
Every sign-up, every card, every budget, watching the meters so a usage bill never surprises anyone.
Which providers, which languages go live first, which towns, how long the acknowledgment timeout is, what counts as ready. The build constantly produces these forks.
The heaviest item. Recruiting testers, running drills, sitting with town operators to compare what the system would do against what they did.
The towns, the responders, and the people who own them. Lumanaire builds; you are the face of it.
POPIA, contracts, liability. Human and slow, and required before a single real caller is allowed near it.
Actual phone calls from actual phones, by actual responders. The "test it for real before trusting it" rule, applied to a phone system.
The software will be ready before the deals and the compliance are. Plan for the deals and the testing to be the critical path, not the code, and start the slow things early.
Lumanaire can build CrisisNexus, and the AI makes that genuinely fast. But the app is not the same as a running emergency service. The service needs paid accounts, partner agreements, privacy compliance, and ongoing management and testing from your side. The build is the fast part. The accounts, deals, compliance and testing are where the time and money go, and for something people's lives touch, that is exactly where they should go.