HRAG – Hybrid RAG on €116/month of Hetzner, officially benchmarked
Source
hrag.app
Author
victor_edka
Date
Context
HRAG describes a document-answering system that keeps documents, keyword search and vector search in one database. It combines retrieved passages and can run an additional model that reads each passage alongside the question to reorder the results.
That extra step illustrates a concrete deployment tradeoff. The developer reports a benchmarkA standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.Full definition → improvement but also 16 seconds of added latency on two virtual CPUs, so rerankingA second pass that re-scores retrieved candidates by reading each one against the query, fixing the ordering that fast vector search got approximately right.Full definition → is optional. The advertised €116 monthly figure covers the hosting cluster; answer-generation charges are separate. This is an account of infrastructure and configuration choices, rather than an all-inclusive price for a production service.
Terms in this piece · Glossary
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
reranking — A second pass that re-scores retrieved candidates by reading each one against the query, fixing the ordering that fast vector search got approximately right.