Blog
Notes and field reports from the team building the Index Server — a private, OpenSearch-compatible search + vector engine you run on the hardware you already own.
MCPagents RAGcuration
Index straight from Claude
September 2026 · 7 min read
RAG problems are usually corpus problems. With MCP, Claude and other agents index content directly into the engine as tools — so the agent that understands your documents also curates them: normalize, tag, dedupe at ingest, then measure with keyword/semantic/hybrid/rerank which sources actually help before committing them to a pipeline. No glue code.
keywordsemantic hybridrerank
One engine, four ways to search
September 2026 · 8 min read
Keyword (BM25), semantic, hybrid (RRF) and reranking are four different retrieval strategies with different strengths — and this engine runs all four over the same index and the same OpenSearch API. What each is good at, where each fails, real latency-vs-quality numbers on a 20k product catalog, and how to pick per use case (including the console's Compare tab).
search saasAlgolia facetsself-host
From Algolia to the Index Server
September 2026 · 8 min read
Algolia is great hosted instant-search; teams leave for self-hosting, data residency, and pricing that doesn't scale with every record and search. Keep your Algolia client via the compatible dialect, or consolidate on the OpenSearch API — with vector + hybrid built in, not a paid add-on. Side-by-side API examples for indexing and faceted search, and where Algolia still wins.
vector dbQdrant hybridconsolidation
From Qdrant to the Index Server
September 2026 · 7 min read
Qdrant is the fastest raw ANN engine in our benchmark — and this post says so. The move is about consolidation: one engine for vector and keyword and hybrid and facets over the OpenSearch API, instead of a vector DB plus a separate search system. Keep your Qdrant client via the compatible dialect; honest numbers and where Qdrant still wins.
vector dbWeaviate OpenSearch APIhybrid
From Weaviate to the Index Server
September 2026 · 7 min read
Both do server-side hybrid; the difference is the API and the surface — drive vector + keyword + hybrid + aggregations over the familiar OpenSearch REST API, with ~2× faster hybrid and ingest. Migrate via the built-in importer; honest numbers, including where Weaviate is the lighter process.
vector dbChroma RAGproduction
From Chroma to the Index Server
September 2026 · 8 min read
Chroma is a great way to prototype RAG; production is where you want server-side hybrid (BM25 + vector), fast filtered vector search, and the operational surface of a real server — auth, TLS, snapshots, HA. Move without re-platforming: keep your Chroma client via the compatible dialect or consolidate on the OpenSearch API. The migration steps, the hybrid-search case, and real Graviton numbers (3 ms filtered kNN vs ~33 ms).
vector dbPinecone costAWS
Replacing Pinecone with the Index Server
September 2026 · 8 min read
A managed vector database bills three ways that all grow with your data: per-vector storage, metered read/write units, and a separate per-token embedding API on every upsert and query. This collapses all three into one server you run — it embeds at ingest for $0 per token, stores vectors on your own EBS/disk, and keeps your Pinecone client working through the compatible dialect. The migration playbook and the AWS cost math.
migrationElasticsearch OpenSearchzero-touch
Migrating your index from Elasticsearch or OpenSearch
September 2026 · 9 min read
Not a re-platforming project: a data copy, a relevance check, and an endpoint swap. The built-in console Migrate tab imports your indexes zero-touch (mappings pass through, counts verified), the Evaluate tab replays your golden queries against both engines side-by-side, and Elasticsearch clients connect via the ES 8 presentation flavor. The full playbook, including the cutover checklist.
scalingclustering raft HAvs OpenSearch & Qdrant
Scaling your AI search with nodes — and how it differs from OpenSearch and Qdrant
September 2026 · 8 min read
AI search corpora are gigabytes; it's the query traffic that grows. Every node here is a full replica of the whole engine — read scaling is one config line, quorum HA with automatic failover is four. Why that's a deliberately different trade than OpenSearch shard clusters or Qdrant distributed mode, and where each model honestly wins.
semantic searchdialects auto-embeddingzero code change
Add semantic search to your AI app without changing your stack
September 2026 · 7 min read
Adding semantic search usually means new infrastructure, a new SDK, and an embedding pipeline. The index server collapses all three: it speaks your existing OpenSearch/Elasticsearch, Qdrant, Pinecone or Chroma dialect, and embeds text server-side — so semantic search becomes a mapping change, not a rewrite.
launchOpenSearch 3.5 drop-inprivate + fixed-cost
Introducing the Index Server: a private, OpenSearch-compatible search engine
September 2026 · 6 min read
A self-hosted search + vector engine that speaks the OpenSearch 3.5 REST API — full-text, kNN and hybrid search, aggregations and facets, native snapshots, and data-plane adapters for Qdrant, Pinecone, Chroma and Algolia. Existing OpenSearch clients connect unchanged. Why we built it, what it does, and where its edges honestly are.