This is not a theoretical stack diagram. This is what actually runs 24/7 on a Hostinger VPS in production, serving multiple SaaS products and an autonomous AI agent.
My Setup
The exact infrastructure behind a one-person AI venture studio
VPS & Infrastructure
Everything runs on a single Hostinger VPS (Ubuntu). No Kubernetes, no multi-region — just a well-configured Linux box.
- Server: Hostinger VPS (4 vCPU, 16GB RAM, 200GB NVMe)
- OS: Ubuntu 22.04 LTS
- Process Manager: PM2 + systemd services
- Reverse Proxy: Nginx with SSL (Let's Encrypt)
- Domains: 5 production domains (adityabiswas.com, creator-os.in, profileinsights.in, qwiklo.in, shopify store)
- Cost: ~$15/month for VPS + ~$60/year for domains
The Tech Stack
Every tool earns its place. No resume-driven development.
- Frontend: Next.js 15 (App Router), React 19, TypeScript, Tailwind CSS + shadcn/ui
- Backend: Supabase (Postgres, Auth, Storage, RLS, Edge Functions)
- AI Models: Google Gemini 2.5 Flash (primary), Gemini 2.5 Pro (deep reasoning)
- Agent Framework: OpenClaw (custom-built, Python-based autonomous agent system)
- Email: Resend (transactional) + SMTP via Google Workspace (newsletters)
- Automation: Python cron jobs managed by OpenClaw scheduler
- Version Control: GitHub (5 repos, all private)
- Social API: Direct X API integration (OAuth 1.0a)
Real AI Costs
What it actually costs to run AI in production
People assume AI is free because ChatGPT has a free tier. Running AI in production is different.
- Gemini API: ~$20-40/month (newsletter generation, blog drafting, content moderation, classification)
- VPS: ~$15/month (runs everything — websites, agent, cron jobs)
- Supabase: Free tier (handles auth, DB, storage for all products)
- Domains: ~$60/year across 5 products
- Google Workspace: ~$6/month (email infrastructure)
- Resend: Free tier (transactional emails)
- Total: ~$50-70/month to run an entire venture studio
The asterisk on "anyone can do this": you need to understand Linux, networking, databases, API design, and deployment pipelines. The AI handles content generation — but the infrastructure that makes it reliable requires engineering.
OpenClaw Agent Architecture
OpenClaw is the autonomous agent system that powers Claw Biswas. It runs as a set of Python services on the VPS.
- Gateway: HTTP API that receives tasks from Mission Control and external triggers
- Scheduler: Cron-based job runner for recurring automations
- Agent Swarm: Specialized sub-agents (Sherlock for intelligence, Writer for content, Scout for monitoring)
- Knowledge Graph: SQLite-based graph tracking entities, themes, and edition history across newsletters
- Memory Engine: Persistent context that lets Claw recall past actions and decisions
- Skills: Modular capabilities (site automation, X API, email, browser automation)
The entire system runs on ~2000 lines of Python. No LangChain, no vector databases, no orchestration frameworks. Just clean scripts, Pydantic schemas, and structured LLM calls.