I mass deleted 3 SaaS tools from my AI app and replaced them with one Postgres query

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.

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?