People assume that running 20 AI agents across 5 projects requires enterprise infrastructure. It does not. Here is the entire stack, what it costs, and why free-tier-first is a strategic advantage, not a compromise.
The Stack
Frontend — React 19 + Next.js + Vite: The AHA website runs Next.js with static export. The Eunoia dashboard runs Vite + React. Both are client-side rendered and deployed as static files. No server. No container. No Lambda functions. Static files on a CDN. The build step is the most complex part, and it runs in CI for free.
Backend — Supabase: PostgreSQL database, authentication, real-time subscriptions, edge functions. The free tier gives you 500MB database, 1GB file storage, 50,000 monthly active users, and unlimited API requests. Andrea has not come close to hitting any of those limits. The paid tier starts at $25/month when she does.
AI — Claude (Anthropic): This is the actual cost center. Claude Pro subscription at $20/month provides the intelligence layer. The agent system does not use the API directly — it operates through Claude Code, Anthropic's CLI tool. System prompts define each agent's identity, scope, and constraints. The agents are not separate API calls; they are structured personas within the same conversation context.
Deployment — Netlify: Free tier. Static site hosting with automatic deploys from GitHub. SSL included. Custom domains included. The AHA website, the Eunoia dashboard, and Kairos all deploy here. Total cost: $0.
Source Control — GitHub: Free tier. Private repos. GitHub Actions for CI/CD (2,000 minutes/month free). That is more than enough for a solo operation.
The Cost Breakdown
Monthly recurring cost:
- Claude Pro: $20
- Supabase Free Tier: $0
- Netlify Free Tier: $0
- GitHub Free Tier: $0
- Domain (amortized): ~$1/month
- Miscellaneous (email, DNS): ~$5/month
Total: roughly $26/month. Call it $50 with buffer for the occasional API overage or paid feature.
That is the infrastructure cost for running a consulting firm with five active projects, twenty AI team members, automated sprint management, and production-deployed products. The entire annual cost is less than one month of most SaaS tool stacks.
Why Free-Tier-First Matters
This is not frugality for its own sake. Free-tier-first is an architectural discipline. When you start with zero infrastructure cost, you are forced into good decisions:
Static over dynamic. If you can serve it as a static file, you should. No servers to maintain. No cold starts. No scaling concerns. The AHA website loads in under 2 seconds globally because it is literally just files on a CDN.
Local-first workflow. Files on disk are the source of truth. GitHub is backup. Supabase is for data that needs to be queryable. This means Andrea can work offline, on a plane, in a cabin with no internet. The AI agents work with local files. The sync happens when it happens.
Simplicity as a feature. Every paid service you add is a dependency you maintain. Every complex architecture is a failure mode you own. When the total stack fits in your head, debugging is fast and outages are rare.
What Scales and What Does Not
This stack scales to roughly 10,000 monthly active users before anything needs to change. At that point, Supabase goes to the Pro tier ($25/month), Netlify might need the Pro tier ($19/month), and Claude API usage may need a direct API subscription instead of Pro.
The agent architecture scales indefinitely because agents are prompts, not processes. Adding a 21st agent costs zero infrastructure. The constraint is context window management, not compute.
For a solo founder, the lesson is clear: do not buy infrastructure for the company you hope to become. Build for the company you are today. Scale when the numbers demand it, not when the anxiety suggests it.