Katalog AI

Katalog AI is a complete, self-hosted multi-tenant SaaS for fashion brands and textile wholesalers. Upload a phone photo of any garment, choose an AI model, select a backdrop — and receive a studio-quality catalogue image in under 60 seconds. Built for Indian SMBs but architected for any fashion business worldwide.

System Requirements

External Services

You need accounts on these platforms before running Katalog AI in production.

Local Installation

git clone <your-repo-url>
cd katalog_ai
pnpm install

cp .env.example .env
# Fill in Firebase, fal.ai, R2, Redis, Postgres credentials

pnpm db:migrate
pnpm db:seed

pnpm dev

Installation

There are two ways to install Katalog AI. Both produce the same running stack — pick whichever matches your comfort level.

git clone <your-repo-url> katalog_ai && cd katalog_ai
pnpm install
make setup

# Open http://localhost:3000 in your browser

cp setup.env infra/docker/.env.katalog
# Edit the file and fill in your real credentials

./deploy.sh up

# Open in browser:
http://localhost:3001   # (local)
https://status.yourdomain.com   # (VPS)

# Choose SQLite → Next → create admin account
# Then seed monitors:
./scripts/seed-uptime-kuma.sh --env prod

Docker Deployment

For VPS deployments (AWS EC2, DigitalOcean, Hetzner). Provides containerised Next.js, worker, Postgres, and Redis.

DATABASE_URL=postgresql://postgres:postgres@katalog-postgres:5432/katalog
REDIS_URL=redis://katalog-redis:6379
FIREBASE_PROJECT_ID=your-project-id
FAL_KEY=fal-xxx

docker compose -f infra/docker/docker-compose.yml \
  --env-file infra/docker/.env.katalog up -d --build

docker compose -f infra/docker/docker-compose.yml exec web pnpm db:migrate

Deploying Chatbots

Both bots are Hono apps deployed to Cloudflare Workers edge.

Four-Step Generation Wizard

The core Web dashboard experience — four streamlined phases from garment photo to studio result.

Custom LoRA Training

Train a dedicated model face by feeding 5–15 reference headshots. The resulting LoRA weight locks face structure across all future shoots — guaranteeing consistent lookbooks regardless of backdrop or pose.

Video Showcase Reels

Convert any generated catalogue image into a 5–15 second cinematic video clip — ready for Instagram Reels, WhatsApp Status, or product pages.

Conversational Bots

WhatsApp and Telegram bring the studio into chat — same AI pipeline, no web browser required.

Billing & Payments

Dual-gateway billing with idempotent webhook processing. Each tenant subscribes to a plan that controls image generation quotas.

Workspace & Library

Each tenant gets an isolated workspace with dedicated sections for every aspect of the generation workflow.

Admin Panel

Super-admin dashboard at /admin for platform operators. Requires the SUPER_ADMIN role.

Monorepo Structure

pnpm workspaces + Turborepo for build caching and isolated execution contexts.

katalog-ai/
├── apps/
│   ├── web/        # Next.js (App Router · RSC · SSE · TypeScript)
│   ├── worker/     # Node.js BullMQ process (fal.ai, face-crop, video jobs)
│   ├── whatsapp/   # Hono + Cloudflare Workers (Evolution webhook)
│   └── telegram/   # Hono + Cloudflare Workers (Telegram Bot API)
│
└── packages/
    ├── db/         # Neon PostgreSQL · Drizzle ORM · Migrations · Seeders
    ├── ai-router/  # Model adapter — fal.ai + Google Vertex AI
    ├── shared/     # Zod schemas · BullMQ queues · storage helpers · errors
    └── config/     # ESLint · TSConfig · Prettier

Async Generation Pipeline

Generative tasks are heavy and async. Katalog prevents HTTP timeouts using a stateful pipeline around BullMQ and webhooks.

System Flowchart

Structural flow from ingest routing through queue validation, worker dispatch, and CDN edge saves.

Database & Row-Level Security

Neon serverless PostgreSQL mapped via Drizzle ORM. Multi-tenant boundaries enforced at the database layer via RLS policies.

Hono Workers & Bot APIs

Both bot integrations are standalone Hono applications deployed to Cloudflare's global edge network.

fal.ai & Vertex AI Integration

Model interfaces are unified inside @katalog/ai-router. This adapter layer maps generation parameters to fal.ai and Google Vertex AI endpoints.

// packages/ai-router — modular adapter interface
import { fal } from "@fal-ai/client";

export async function dispatchImageGeneration(payload: ImageJobPayload) {
  return await fal.queue.submit("fal-ai/flux-lora", {
    input: {
      prompt: payload.structuredPrompt,
      image_size: payload.aspectRatio,
      loras: payload.customLoraUrl
        ? [{ path: payload.customLoraUrl, scale: 0.85 }]
        : [],
      num_inference_steps: 28,
      enable_safety_checker: true,
    },
    webhookUrl: `${process.env.FAL_WEBHOOK_BASE_URL}?jobId=${payload.jobId}`,
  });
}

Security Hardening

Production hardening applied as part of Phase 12 launch preparation.

Environment Variables

Copy .env.example at root and fill in all keys before deploying.

FAQ & Troubleshooting