Building Kataru on AWS for Under €300/Month: Our Infrastructure Stack

One of the commitments we made when starting Kataru was to build in public where we could. That means showing not just the product decisions, but the engineering ones too — including the parts that involve money.

This is a snapshot of our infrastructure stack as of May 2026. We’re in alpha, with a small user base and no external funding, so keeping costs predictable matters.

The stack

Compute: ECS Fargate (FastAPI backend, containerised). No EC2 to manage, no idle instances. We pay per task-second of actual CPU and memory usage.

Database: RDS Postgres (db.t4g.micro in alpha, with pgvector extension for embedding search). Single-AZ for now — not production-grade for a scaled product, but appropriate for alpha.

Storage: S3 with KMS encryption for user documents. Lifecycle rules archive anything older than 90 days.

CDN / static hosting: CloudFront + S3 for the marketing site. Zero server cost for the landing page.

Auth: AWS Cognito for user management and token issuance. Free tier covers 50,000 MAU, which we won’t touch for a long time.

What it costs

As of alpha launch (single-digit daily active users):

Service	Monthly est.
ECS Fargate (1 task, 0.25 vCPU / 512MB)	~€6
RDS db.t4g.micro	~€12
CloudFront + S3 (marketing)	~€2
Cognito	€0 (free tier)
Secrets Manager, KMS	~€2
Total	~€22

With real user traffic, the first expensive line item will be Fargate tasks (we’ll scale to more than one) and RDS (we’ll need Multi-AZ). At 1,000 DAU, we’d estimate €80–120/month — still well under the target.

What we traded off

We chose Terraform for all infrastructure, which added upfront time but gives us full auditability and reproducibility. Every resource is defined in code, versioned in git, and deployable from scratch in under 15 minutes.

We skipped managed services where the cost-to-value ratio was poor at this scale — no ElasticSearch, no SQS, no EventBridge for MVP workflows. We use Postgres for a lot of what those services would do, which is a sensible trade for a two-person team.

What’s next

The next infrastructure post will cover our CI/CD pipeline — GitHub Actions deploying to ECS, with zero-downtime rolling updates and canary health checks.

If you’re building something similar and want to compare notes, reach out on LinkedIn.