{"slug": "show-hn-hrag-hybrid-rag-on-eur116-month-of-hetzner-officially-benchmarked", "title": "Show HN: HRAG – Hybrid RAG on €116/month of Hetzner, officially benchmarked", "summary": "A developer-built hybrid retrieval-augmented generation (RAG) platform, hRAG, ranks #9 on the EnterpriseRAG-Bench leaderboard with an overall score of 44.74, running entirely on a €116-per-month Hetzner cluster. The system, built by Victor Mocioiu, achieves 88ms BM25 search over 2 million chunks using Postgres and a cross-encoder reranker, and is open-sourced on GitHub.", "body_md": "[Read Part 1 ↗](https://medium.com/@victormocioiu/production-grade-hybrid-search-rag-on-116-month-of-compute-part-1-the-pipeline-the-deede53d461d)·\n\n[GitHub](https://github.com/victormocioiu/rag-api)\n\n# hRAG\n\nHybrid retrieval with receipts.\n\n**#9**\n\n[on EnterpriseRAG-Bench — officially scored](https://huggingface.co/spaces/onyx-dot-app/EnterpriseRAG-Bench-Leaderboard)\n\n**512K** docs benchmarked\n\n**88ms** BM25 at 2M chunks\n\n**€116** a month of Hetzner… yes, that is the whole cluster\n\n00 · 01\n\n## The platform, answering\n\nactual screenshots of the live thing… the 512K-document playground, streamed answers, citations you can open. no mockups — mockups would have been easier.\n\n01 · 02\n\n## What runs under the hood\n\n-\nBM25 inside PostgresNo search cluster to babysit.\n\npg_textsearch gives real IDF ranking with Block-Max WAND: 88ms over 2 million chunks, where plain tsquery took 7.5 seconds.\n\n-\nHybrid fusion, weightedFusion is not a democracy.\n\nVector and lexical arms fuse with a measured 0.3 vector weight — equal weighting scored worse than turning one arm off.\n\n-\nCross-encoder rerankerThe right eight chunks.\n\nA fourth service reads query and chunk together and reorders the window. Worth +3.9 on the public benchmark… and 16 seconds of latency on 2 vCPUs, which is why it ships as a checkbox and not a default. Both numbers are published.\n\n-\nGrounded answersRefusal as a feature.\n\nEmpty retrieval refuses without calling the model. On the benchmark's info-not-found questions: 100% correct while others hallucinate.\n\n-\nTenants by row-level securityIsolation is the floor, not a filter.\n\nEvery query runs inside a tenant transaction Postgres enforces. The benchmark corpus and your sandbox are neighbors that cannot meet.\n\n-\nStreaming citations[n] means something.\n\nSources arrive before the first token; every bracket in an answer opens to the exact chunk that backs it.\n\n02 · 03\n\n## How the numbers stay honest\n\n- 1\n**Measure before building** The reranker cost $1 to prototype offline before a line of service code existed. - 2\n**Publish the noes** Multi-query rewriting, wider context, prompt-only fixes, HyDE — measured, failed, documented next to the wins. The noes cost real money; hiding them would waste it twice. - 3\n**One database** Vectors, BM25, text, tenants, and usage all live in Postgres. One thing to operate, one thing to trust. - 4\n**Small models, priced honestly** A 118M embedder and budget answerers, benchmarked against clouds that cost 100× more.\n\n03 · 04\n\n## What it costs\n\n- Five-node Kubernetes cluster on Hetzner\n- Postgres with vectors, BM25, and row-level security\n- Four services: ingest, embed, rerank, answer\n- A public benchmark score you can reproduce\n- All code MIT, all numbers receipted\n\n€116\n\nAnswers cost tenths of a cent; the benchmark that proved them cost about $60 once.\n\n[Try it now — no login](/app)\n\nThe 512K-document playground is open to everyone, no login. Sign in with Google or GitHub only for your own private sandbox.\n\nper month — the entire five-node cluster\n\n04 · 05\n\n## Fair questions\n\n- What am I actually chatting with?\n- EnterpriseRAG-Bench: 512,000 simulated company documents — Slack threads, emails, wikis, tickets… the same corpus the platform is publicly scored on. what you see is exactly what the benchmark saw.\n- Can I bring my own documents?\n- Yes — sign in with Google or GitHub and you get a private sandbox: 10 documents, 20 pages each, a daily token budget, and meters that show exactly what's left. A page is counted in tokens, so no… one huge .txt file is not \"one page\", we checked. Row-level security keeps your docs yours.\n- How good is it, honestly?\n- Officially scored on EnterpriseRAG-Bench:\n**#9**, overall 44.74 (correctness 52.6, completeness 54.38, document recall 69.65) — ahead of Vertex AI Search and NVIDIA, one spot below Azure. On document recall we beat both Azure (64.25) and Vertex (61.76): the retrieval competes with the managed clouds outright, on a €116/month box. The[live leaderboard is on Hugging Face](https://huggingface.co/spaces/onyx-dot-app/EnterpriseRAG-Bench-Leaderboard); per-category tables are in the repo. - Why should I trust these numbers?\n- You shouldn't — not on our word alone. Every run's raw results are committed to the repo, and when the benchmark team re-scored our submission with their own judge, their recall landed within 0.05 points of ours (69.65 vs 69.6). Retrieval validated to a rounding error. Verify, then trust.\n- Can I run this myself?\n- That is the whole point… the articles walk through every deployment click and every benchmark, and the repos build the platform from scratch. Copy freely, measure everything.", "url": "https://wpnews.pro/news/show-hn-hrag-hybrid-rag-on-eur116-month-of-hetzner-officially-benchmarked", "canonical_source": "https://hrag.app/", "published_at": "2026-08-19 17:36:01+00:00", "updated_at": "2026-08-19 17:59:42.075672+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "generative-ai", "ai-infrastructure", "ai-products"], "entities": ["hRAG", "Victor Mocioiu", "Hetzner", "EnterpriseRAG-Bench", "Postgres", "GitHub", "Vertex AI Search", "Azure"], "alternates": {"html": "https://wpnews.pro/news/show-hn-hrag-hybrid-rag-on-eur116-month-of-hetzner-officially-benchmarked", "markdown": "https://wpnews.pro/news/show-hn-hrag-hybrid-rag-on-eur116-month-of-hetzner-officially-benchmarked.md", "text": "https://wpnews.pro/news/show-hn-hrag-hybrid-rag-on-eur116-month-of-hetzner-officially-benchmarked.txt", "jsonld": "https://wpnews.pro/news/show-hn-hrag-hybrid-rag-on-eur116-month-of-hetzner-officially-benchmarked.jsonld"}}