Rivestack

I'm a DevOps engineer by day. I spend most of my time managing Kubernetes clusters and CI/CD pipelines for a large org. But I've been building stuff on the side for a while, and a few weeks ago I launched Rivestack — a managed PostgreSQL service with pgvector for AI workloads.

The problem is obvious: nobody knows it exists.

I don't have a marketing budget. I don't have an audience. I don't even have a Twitter following worth mentioning. So I needed to figure out how to get my first users without spending money I don't have.

The idea

I kept coming back to one thought: what if I just built something useful and gave it away?

Hacker News search has always been... okay. It works for exact keywords but it's frustrating when you're looking for a concept. Like, I wanted to find old discussions about people moving from dedicated vector databases back to Postgres, but I couldn't remember the exact phrasing. Keyword search doesn't help with that.

So I built https://ask.rivestack.io — semantic search over Hacker News. You type a question in natural language, and it finds relevant threads by meaning, not just matching words.

The whole thing runs on Rivestack itself. One Postgres instance with pgvector, some embeddings, done.

Why this approach

I could have written blog posts. I could have cold-DMed people on Twitter. I could have posted "check out my SaaS" on Reddit and watched it get downvoted into oblivion.

Instead I wanted to make something that the exact people I'm trying to reach would actually want to use. Developers who hang out on HN are exactly the kind of people who might need a managed Postgres with vector search. And if the tool works well, they don't need me to explain what Rivestack does — they can see it.

It's not a genius strategy. It's literally just: make a useful thing, put your name on it, hope people notice.

What it cost me

Honestly, not much. The hardest part was indexing HN data and generating embeddings. The infra cost is minimal because it's running on my own platform. I spent maybe a weekend on it.

The trade-off I keep thinking about is: is this better than spending that weekend writing 5 blog posts? Probably yes, because a working demo says more than any blog post ever could. Nobody reads "we deliver sub-4ms vector search latency" and gets excited. But when you type a vague question and get a perfect result instantly — that clicks.

What Rivestack actually is

Since you're on Indie Hackers, here's the business side. Rivestack is managed PostgreSQL with pgvector built in. You get a database in minutes, with backups, metrics, SSL, autoscaling. Free tier for testing, paid plans start at $29/month per node.

The bet I'm making is that most teams building AI apps don't need Pinecone or Weaviate or Qdrant. They already use Postgres. They just need Postgres with good vector search and someone else handling the ops. That's it.

I'm a one-person team running this alongside my day job, so I'm not trying to compete with Neon or Supabase on features. I'm trying to be the simplest option for people who just want pgvector to work without thinking about it.

What's next

I'm going to post the semantic search tool on HN as a Show HN and see what happens. Maybe it gets 3 upvotes and disappears. Maybe someone finds it useful. Either way, I'll learn something.

If you have thoughts on better ways to get early users when you have zero audience and zero budget, I'd genuinely love to hear them. I'm figuring this out as I go.

You can try the search at https://ask.rivestack.io and the platform at https://rivestack.io.

I was building an AI app and my stack looked like this:

- Pinecone for vectors ($70/mo)

- Supabase for user data ($25/mo)

- Separate backup service ($10/mo)

- Redis for caching metadata ($15/mo)

Total: ~$120/mo just for data storage. For a side project.

Then I discovered pgvector and realized I could do it all in Postgres:

- Vectors? pgvector extension

- User data? It's Postgres

- Backups? Built in

- Metadata? Just another table

One query to find similar items + join user data + filter by date:

SELECT p.*, u.name FROM products p JOIN users u ON p.user_id = u.id ORDER BY p.embedding <-> $1 LIMIT 10;

Try doing that with Pinecone + Supabase + Redis.

Problem was: managed Postgres with pgvector is either expensive (AWS = $150+) or shared resources (Neon free tier).

So I built Rivestack.

Dedicated PostgreSQL with pgvector. $29/month. Free tier to test.

Benchmarks on the $29 tier:

- 10k vectors: 2,000 QPS, <4ms

- 1M vectors: 252 QPS, 98% recall

I just shipped it to production after months of building.

🔗 Try the demo: https://ask.rivestack.io

🔗 Landing page: https://rivestack.io

Looking for 10 early users who are building AI/RAG apps. Happy to give free Pro tier for 3 months in exchange for honest feedback.

Anyone else frustrated with the "you need 5 services" approach to building AI apps?