LolaDesk — AI Front Desk for Salons & Spas

A multi-tenant SaaS where a salon owner runs their entire front desk through Lola — an AI that answers calls (in her real, custom ElevenLabs voice), texts, books appointments, and surfaces revenue opportunities. Built to be sold direct and listed on Square, Mindbody, Vagaro, and Fresha.

Architecture — what's actually real

This is not a static demo. Every salon is a row in Supabase, isolated by Row Level Security, with their own phone number, services, persona, integrations, and conversation history.

Browser (static HTML/CSS/JS, no build step)
   ↓
Vercel serverless functions (api/*.js)
   ↓
Supabase (Postgres + Auth)     Telnyx (voice + SMS + numbers)     ElevenLabs (Lola's voice)
   ↓                                ↓                                  ↓
tenants · clients · conversations   inbound calls/texts route to    every spoken word, on
· messages · calls · bookings ·     /api/telnyx-voice and           calls AND in the
usage_events · integrations         /api/telnyx-sms by called #     dashboard, same voice

Pages

File	Purpose
index.html	Marketing homepage — what loladesk.com serves at /.
onboarding.html	Setup wizard — salon details → connect platform → services → Lola's personality → plan → live. Calls /api/auth/signup and /api/onboard.
dashboard.html	The dashboard — what each salon logs into daily. Reads real data via /api/data. Lola can be talked to two ways: tap-to-talk, or enable "always listening for Lola" for hands-free, ambient wake-word activation (mic stays passively open; nothing is sent to the AI brain until her name is said).
clients.html, calls.html, inbox.html, bookings.html, revenue.html, team.html, numbers.html, marketing.html, settings.html, agents.html	Interior dashboard pages, each tenant-scoped via /api/data.
login.html	Returning owner sign-in.
book.html	Public booking page — the same engine Lola uses on the phone, for clients who book on the web. Any URL ?t=SLUG works; it's a thin host for booking-widget.js.
landing.html	Alternate marketing page (not currently linked from main nav).

The API layer (api/)

Path	What it does
auth/signup.js, auth/login.js, auth/session.js	Supabase Auth — creates the owner's account + their tenant row + 14-day trial.
onboard.js	Saves tenant details; if a website URL is given, the Marketer agent analyzes it and pre-fills services/positioning.
telnyx-voice.js	Real phone calls. Telnyx hits this on every call. Resolves tenant by called number, asks Lola's brain, speaks the reply in her real ElevenLabs voice via <Play> (falls back to a generic <Say> only if ElevenLabs fails), persists the conversation.
telnyx-sms.js	Real texts — SMS and WhatsApp. Same tenant resolution + memory, plus STOP/HELP/START keyword compliance (10DLC) handled before Lola's AI ever sees the message. Telnyx WhatsApp payloads (type: WHATSAPP) route through this exact pipeline; webhooks/whatsapp.js is a thin alias onto it.
telnyx-numbers.js	Search & buy phone numbers; auto-attaches voice + SMS on purchase. This is a recurring revenue line (see TELNYX-SETUP.md).
telnyx-agents.js	Experimental, not yet wired to live calls. Provisions a 7-agent team (Lola + 6 specialists) directly inside Telnyx's own AI Assistant product, as an alternative architecture to the custom TeXML+Claude loop above. No phone number currently routes to it — agents.html's "copy Telnyx config" button exports this for manual import if you want to experiment with it.
lola-tools.js	The skill layer (check availability, book, quote pricing, capture leads, escalate) — what Lola's brain calls into to actually do things.
lola.js	Proxy the dashboard's browser-side chat uses to talk to the LLM without exposing API keys.
speak.js, lib/elevenlabs.js	Lola's real voice — shared between the dashboard chat and phone calls so she sounds identical everywhere. The dashboard's app.js calls speak.js for both tap-to-talk replies and ambient wake-word replies (see dashboard.html's "always listening for Lola" toggle).
voice-audio.js, lib/tts-cache.js	Serves the synthesized ElevenLabs audio at a URL Telnyx's can fetch.

Run it locally

python3 -m http.server 8080
# open http://localhost:8080/index.html — the marketing site
# open http://localhost:8080/dashboard.html — straight to the dashboard (demo data without env vars)

The api/*.js files are Vercel serverless functions and won't run under the plain Python server — use vercel dev for those, or just deploy to a Vercel preview to test the full stack.

Open-source booking — embed it on any website

The booking system is open and embeddable. Any salon — on LolaDesk or not — can put Lola's booking flow on their own site with one script tag. It uses the same /api/public-booking engine as her voice, so web, phone, and dashboard bookings share one conflict-free calendar.

<!-- Inline flow (service → team → time → details → confirmed) -->
<script src="https://www.loladesk.com/booking-widget.js"
        data-tenant="YOUR-SLUG"
        data-accent="#ccff00"
        data-mode="inline"></script>

<!-- Or modal mode: a floating "Book now" button opens the flow -->
<script src="https://www.loladesk.com/booking-widget.js" data-tenant="YOUR-SLUG" data-mode="modal"></script>

The widget is a single file (booking-widget.js), vanilla JS, Shadow DOM (never fights the host site's CSS), no build step. book.html is the canonical standalone host. Sync works the other way too — LolaDesk ingests any salon's existing booking system via the connector contract. See CONNECTORS.md for how to add Square/Vagaro/Mindbody/Boulevard/Fresha/Google Calendar — or any new system — in three steps, with the conformance suite in tests/connector-contract.test.mjs.

Deploy

See DEPLOY.md for the Git → Vercel flow, SUPABASE-SETUP.md for the database, TELNYX-SETUP.md for voice/SMS, and .env.example for the full list of environment variables this code actually reads — keep that file in sync if you add a new process.env.X anywhere.

Multi-tenant: one codebase, every salon

Each salon is a row in the tenants table (see schema.sql), identified by the phone number they're assigned. Every inbound call or text resolves its tenant by the called number, loads that salon's services/persona/hours/integrations, and Lola responds accordingly — fully isolated from every other salon via Postgres Row Level Security plus tenant-scoped queries in application code.

To onboard a new salon: they sign up (onboarding.html → /api/auth/signup), pick a number (numbers.html → /api/telnyx-numbers), and Lola is live on that number immediately — no code changes, no redeploys.

Brand

• Fonts: Inter + Cormorant Garamond (Google Fonts, free)
• Icons: inline SVG (no dependency)
• Voice: one consistent ElevenLabs voice for Lola everywhere — dashboard chat and real phone calls alike (see api/lib/elevenlabs.js)