Same substrate. Any vertical.
Each vertical is a 20-line configuration on top of the identical FallHub core. Different kernel (what the AI knows). Different tools (what the AI can do). Different personas (who your customers are). Same event log, same autonomy dial, same training simulator, same module marketplace.
Pick a template · configure · deploy.
Hospitality
Trades
Ad firm
Accounting practice
Barbershop chain
Your niche · bespoke
How to build a vertical for any small business.
A vertical template is three files on top of the FallHub core. Total code: usually under 300 lines. Total time from spec to live deployment: 3-5 days for a new domain, one afternoon for a variant.
1 · The kernel
A JSON file that describes what the business is. Facts the AI must respect. Facts the AI must never invent.
// vertical kernel { "business": { "name": "Rivendell Restaurant", "type": "restaurant", "seats": 42, "cuisine": "seasonal british", "not_present": [ "outdoor seating", "private dining room" ] }, "voice": { "tone": "warm" }, "policies": { ... } }
2 · The tools
A registry of verbs the AI is allowed to invoke. Each verb has a JSON schema (for function-calling), guardrails, and a reversibility window.
// tool registry { "name": "take_reservation", "parameters": { ... }, "guardrails": { "auto_low": true, "max_party": 8 }, "reversible_ms": 30000, "execute": async (args) => { // write to calendar } }
3 · The personas
Archetypes of your typical customers plus scenarios they trigger. Powers the training simulator so the human operator can practice before real customers arrive.
// sim personas + scenarios [ { "name": "Anna Mercer", "archetype": "dietary_restriction", "traits": "coeliac, specific", "temperament": 0.7 }, ... ] // scenarios: reservation, dietary_ask, complaint, walk_in, group_booking...
Not sure what your niche needs?
Book a Pro setup and we'll build your vertical template for you. You keep the code (MIT). We publish it back to the estate so future SMBs in the same niche inherit it. Everyone wins.
What you get regardless of vertical.
- BYOK adapter across 6 LLM providers (Anthropic, OpenAI, Gemini, Groq, Mistral, WebLLM local)
- Kernel-grounded reasoning so the AI can't invent facts about your business
- Ed25519-signed event log for audit trail and reversibility
- Autonomy dial from Watch → Suggest → Auto-Low → Auto-Med → Full
- Training simulator with WebLLM personas so you practice before going live
- 60+ estate modules from a marketplace of pre-built capabilities
- Data portability · one-click signed JSON export of everything
- MCP exposure so external agents (Claude Code, custom bots) can drive your instance
- MIT licence · own it, fork it, run it forever · no lock-in