PropFlow

AI-Powered Real Estate Sales CRM

PropFlow is a B2B SaaS platform for real estate agencies and sales teams. It helps organizations capture, qualify, and prioritize leads; manage sales pipelines, properties, follow-ups, and appointments; and use AI for lead scoring, summarization, and sales intelligence.

Status: Checkpoint 1 (application shell) complete. See docs/checkpoints.md for the full build log.

Features

• Lead capture, qualification, scoring, and assignment
• Kanban sales pipeline (New -> Contacted -> Qualified -> Property Matched -> Site Visit -> Negotiation -> Won / Lost)
• Property inventory and rule-based lead-to-property matching
• Follow-up and appointment scheduling
• Deal tracking and revenue reporting
• AI lead scoring, lead summaries, and draft follow-up messages (human review required before sending - nothing is sent automatically)
• CRM-aware AI assistant for natural-language questions about the pipeline
• Lead, sales, team, and property analytics
• Multi-currency, multi-timezone, organization-level settings
• Role-based access for Admin, Manager, and Agent, enforced in both the UI and the database (Row Level Security)

Architecture

app/
  (app)/              route group for every authenticated module,
                       wrapped in the shared AppShell
    dashboard/
    leads/
    pipeline/
    properties/
    follow-ups/
    appointments/
    ai-assistant/
    analytics/
    team/
    settings/
    help/
  layout.tsx           root layout: fonts, metadata
  page.tsx              redirects "/" to "/dashboard"
  globals.css           design tokens (colors, radius, shadows, type)

components/
  ui/                  reusable primitives (Button, Badge, Card, Avatar,
                       Input, Skeleton, EmptyState)
  layout/              Sidebar, Topbar, MobileNav, AppShell, PageHeader,
                       ModulePlaceholder

lib/
  utils.ts              cn() class-merge helper
  nav-config.ts         single source of truth for sidebar navigation

docs/
  design-system.md      token reference
  checkpoints.md         build log, updated at the end of every checkpoint

Server components are used by default; anything interactive (the sidebar's active-route highlighting, the mobile nav drawer, the topbar) is marked "use client" explicitly and kept as small as possible.

Technology stack

Layer	Choice
Frontend	Next.js (App Router), TypeScript
Styling	Tailwind CSS v4, hand-rolled shadcn-style primitives (class-variance-authority, tailwind-merge)
Fonts	Geist Sans / Geist Mono, self-hosted via the geist npm package
Backend	Supabase (PostgreSQL, Auth, Storage, Row Level Security)
AI	Server-side only - API key never exposed to the client
Deployment	Vercel

Getting started

npm install
cp .env.example .env.local   # fill in Supabase + AI credentials
npm run dev

The app currently redirects / -> /dashboard. Every module route renders as a labeled placeholder until its checkpoint lands (see docs/checkpoints.md) - this is expected; no functionality has been wired up yet beyond navigation and layout.

Environment variables

See .env.example.

Variable	Where it's used	Notes
NEXT_PUBLIC_SUPABASE_URL	client + server	Public - safe to expose
NEXT_PUBLIC_SUPABASE_PUBLISHABLE_KEY	client + server	Public - safe to expose, RLS enforces access control
AI_API_KEY	server only	Never referenced with the NEXT_PUBLIC_ prefix; all AI calls happen in server actions / route handlers

Database

PropFlow uses an existing, already-provisioned Supabase project as the source of truth. The application code does not create, drop, or rename tables, and does not weaken Row Level Security. See docs/checkpoints.md for exactly which checkpoint introduces the Supabase client and generated types.

Core tables (already provisioned, tenant-isolated by organization_id via RLS): organizations, profiles, leads, properties, lead_properties, activities, follow_ups, appointments, deals, ai_analyses.

Tenant isolation model: authenticated user -> profile -> organization_id -> records scoped to that organization.

Security

• No secret keys are ever placed in client-side code or NEXT_PUBLIC_ variables.
• Privileged operations (AI calls, cross-tenant admin actions) run server-side only.
• Role permissions (Admin / Manager / Agent) are enforced at the UI layer and, from Checkpoint 2 onward, at the database layer via RLS - the UI is never the only gate.
• Organization IDs are never trusted from client input for authorization decisions.

AI features (planned - Checkpoints 6-7)

• Lead scoring - 0-100 score with priority (cold/warm/hot) and a plain-language rationale, persisted to ai_analyses (analysis_type = lead_scoring).
• Lead summary - on-demand natural-language summary of a lead's requirements and activity.
• Follow-up message drafting - generates an email or WhatsApp-style draft; a human always reviews and sends it manually, nothing is auto-sent.
• AI Assistant - answers natural-language questions about the pipeline using real application data, with links back to the records referenced.

Deployment

Target platform is Vercel. Set the environment variables above in the Vercel project settings before deploying; nothing in this repo assumes a specific hosting provider beyond that.

Demo data

Seed/demo data (organization: PrimeNest Realty; example agents and leads) will be clearly labeled as demo data wherever it appears in the UI - it is never presented as if it were a real customer's data.