Back

AgentBox

AI Agent Platform

Visit site ↗
$

The Problem

Companies need AI agents but building them requires deep technical knowledge. Existing platforms are either too simple (no multi-step pipelines, no integrations) or too complex (require ML engineers). There was no middle ground for business users who need production-grade AI agents with proper security and multi-tenant isolation.

$

The Solution

Built a full SaaS platform where companies can create, configure, and deploy AI agents without writing code. Each company gets isolated data, their own agent library, and can connect channels like WhatsApp and Slack. The platform handles prompt injection prevention, rate limiting, and document generation automatically.

$

Technical Decisions

> Supabase over Firebase — needed Row Level Security for true multi-tenant isolation at the database level, not just application-level checks

> Gemini as primary LLM — best cost-to-quality ratio for high-volume agent workloads, with Claude and OpenAI as fallbacks

> Server-side document generation — Word/PDF/Excel generated on the server to avoid exposing templates and business logic to the client

> Twilio for messaging — single provider for WhatsApp, SMS, and voice with reliable webhook delivery and global coverage

$

Stack Breakdown

Frontend

Next.jsReactTailwind CSS

App Router with SSR, shadcn/ui components

Backend

Next.js API RoutesSupabase Edge Functions

144 routes handling agents, pipelines, integrations

Database

Supabase (PostgreSQL)

62 migrations, RLS policies on every table

AI

GeminiClaudeOpenAI

Multi-provider with automatic fallback

Infra

VercelCloudflare

14 cron jobs, edge caching, DDoS protection

$

Numbers

144

routes

62

migrations

14

crons

Visual coming soon