# I built a vector database hosted in Mumbai because querying US servers from India was adding 200ms+ latency — retained record

> Retained reddit record from reddit:developersIndia.

Canonical: https://highsignal.app/data/records/e45beebe95fbbb3a

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:developersIndia)  published 2026-10-24 05:28 ingested 2026-10-24 10:57

[original source ↗](https://www.reddit.com/r/developersIndia/comments/1xe0asf/i_built_a_vector_database_hosted_in_mumbai/)

## retained text



```
Hey everyone,



Over the past few months I have been working on a solo project called Iceberg DB (https://icebergdb.in) and wanted to share it here to get some honest feedback from other builders.



The main reason I started building this was frustration with existing cloud vector databases while working on RAG and AI agent projects from India. Almost every managed vector DB hosts its free or starter tiers in US-East or Europe, which adds around 180-250ms of network round-trip time before the vector search even happens. On top of that, paying $50+/month in USD plus separate OpenAI API bills just for embeddings gets expensive pretty quickly for indie devs and early startups.



Here is what I built into Iceberg DB so far:



\- Hosted in Mumbai (ap-south-1), so query latency inside India stays under 20ms.

\- Built-in server-side embeddings using FastEmbed (384-dim MiniLM). You can just send raw text or upload a PDF directly and it chunks and embeds it on the server without needing an OpenAI API key.

\- Hybrid search out of the box (combines dense vector similarity + BM25 keyword search using RRF) so exact product codes or names don't get missed.

\- A built-in agent memory endpoint (/memory) with short-term (1h TTL), episodic (30d TTL), long-term, and semantic memory so LLM agents can recall user context across sessions.

\- SDKs for Python (pip install iceberg-db), JS/TS, Go, Rust, C# (.NET), and Java.



There is a free tier with 100k vectors that doesn't pause on inactivity, and no credit card is needed to get an API key.



Link: [https://icebergdb.in](https://icebergdb.in)



If you are building anything with RAG or AI agents, I would genuinely love for you to try breaking it and tell me what feels clunky, what is missing, or how I can improve the docs/SDKs. Happy to answer any questions about the backend stack or architecture!
```



document fetched 2026-10-25 00:24 · canonical https://reddit.com/r/developersIndia/comments/1xe0asf/i_built_a_vector_database_hosted_in_mumbai
