How to not go broke building an AI agents startup

So here's the thing : building AI apps is the fun part. Deploying them? That's where your soul goes to die.

You've got two options and they both suck:

Self-host on a VPS → congrats, you're now a DevOps engineer. What's "kubirnites"?

Go serverless → hit execution limits when your AI agent is mid-thought

Vercel caps functions at ~500 seconds even with fluid compute. Cool, except my LangGraph agents take more than 500 seconds depending on the workflow.

So here is my 0$ stack to run my ai agents SaaS:

I realized GitHub Actions workflows can run for free, up to 2,000 minutes/month on the free tier (6 hours for private repos). So I just... moved all my long-running AI stuff there.

Load LangGraph script → trigger via .yml → dispatch through an endpoint → webhook the results back. Async, stable, and GitHub's problem now, not mine.

The rest of my $0 stack:

Frontend: SvelteKit (a JS framework that doesn't make me want to cry or go into wrapper hell)

Backend: Django for auth (Learned it and stuck with it)

DB: Neon serverless Postgres (the best postgres db in the world. practically free)

AI: Gemini with Google Cloud free credits

Storage: S3 free tier

Analytics: PostHog free tier

Hosting: Vercel (just frontend + lightweight APIs)

The heavy lifting: GitHub Actions

This powers my SaaS my SEO automation tool. Been running for months. Total monthly cost: $0. and i am pretty happy with it overall

So is this the best way to run Ai agents for free? Is anyone else doing cursed infrastructure hacks like this, or am I the only one?

For those running production AI products what's your deployment look like? Are you just eating the compute costs or is there a better way I'm missing?