Search Platforms
Overview
A search platform is the engine that stores, indexes, and retrieves content at query time. Choosing a platform is one of the earliest and most consequential infrastructure decisions for a search team — it determines what retrieval methods are available, how much operational overhead the team carries, and what the cost structure looks like at scale.
Platforms split along two axes:
- Open source vs. proprietary — open-source engines give full control over the stack and data; proprietary SaaS solutions trade that control for faster onboarding, managed infrastructure, and vendor-provided relevance tooling.
- Self-hosted vs. SaaS — even open-source engines (e.g. Elasticsearch, OpenSearch) are available as managed cloud services; pure SaaS products (Algolia, Pinecone) are only available hosted.
Lucene-Based
Engines built on Apache Lucene. Lucene provides a full-featured foundation: inverted index and BM25 for lexical retrieval, but also native support for vector fields (HNSW) and hybrid search. These engines are not limited to keyword search — they are increasingly used as unified retrieval backends.
- Elasticsearch — the dominant choice for production search; rich query DSL, extensive ecosystem. Available open-source (self-hosted) and as Elastic Cloud (SaaS).
- Apache Solr — the other major Lucene-based engine; older, more configuration-heavy, common in enterprise and media.
- OpenSearch — AWS-led fork of Elasticsearch (post-7.x licensing split); fully open-source, available managed on AWS.
Vector / Semantic Search
Purpose-built for dense embedding retrieval. These engines index high-dimensional vectors (via HNSW or similar ANN structures) and return results by similarity rather than keyword overlap. Most support hybrid search combining keyword and vector scores.
- Qdrant Vector DB — open-source, Rust-based, strong filtering performance; available self-hosted and as Qdrant Cloud (SaaS).
- Weaviate Vector DB — open-source, GraphQL API, built-in vectorization modules; available self-hosted and as Weaviate Cloud (SaaS).
- Pinecone — proprietary SaaS-only vector database; no self-hosting option; known for operational simplicity.
- Milvus / Zilliz — open-source vector DB (Milvus) with a managed SaaS layer (Zilliz Cloud).
- Chroma — lightweight open-source vector store popular in RAG prototyping.
Full-Featured / Hybrid Platforms
Engines designed from the ground up with relevance, hybrid retrieval, and production scale as first-class concerns.
- Vespa — open-source, self-hosted (or Vespa Cloud SaaS); uniquely combines BM25, dense vector, and structured attribute matching in a single engine via a general tensor ranking-expression framework. Strong choice for complex ranking pipelines — it expresses ONNX LTR and late-interaction MaxSim as ranking expressions rather than bolt-on subsystems. Production ES→Vespa migrations: Vinted - Migrating Search from Elasticsearch to Vespa (billion-item scale) and Kleinanzeigen - Vespa Migration for Homepage Feed.
- Typesense — open-source, self-hosted or Typesense Cloud; developer-friendly, fast, opinionated defaults; positioned as a simpler alternative to Elasticsearch for search-heavy apps.
- Meilisearch — open-source, self-hosted or Meilisearch Cloud; optimized for instant search UX, typo tolerance, and ease of setup.
SaaS / Proprietary Platforms
Fully managed, proprietary offerings where the engine is not available for self-hosting.
- Algolia — the dominant SaaS search platform for product and site search; strong relevance-tuning UI, extensive front-end SDKs, high price point at scale.
- Coveo — enterprise SaaS search with ML-based ranking and personalization; targets large organizations.
Specialized / Vertical Platforms
Some platforms are not general-purpose search engines but focus on a specific domain or problem layer, typically sitting on top of an underlying engine.
- searchHub (Andreas Wagner) — e-commerce-focused SaaS platform for search optimization and query understanding. Covers query rewriting, A/B testing, analytics, and spell correction (SmartQuery). Also released the Open Commerce Search Stack (OCSS) as an open-source reference implementation for retail search. See Three Pillars of Search Quality - Findability and Common Pitfalls of Onsite Search Experimentation.
- Hornet (Jo Kristian Bergum, Lester Solbakken) — retrieval infrastructure platform purpose-built for AI agents. Focuses on high-precision context delivery for agentic workloads, where query patterns differ fundamentally from human search (longer queries, web-search operators, multi-turn compounding). See This Is What Agentic Retrieval Looks Like and Mutually Assured Distraction.
- Empathy.co — e-commerce search and discovery platform; SaaS, focused on search UX, analytics, and personalization.
- Constructor.io — SaaS product discovery platform for e-commerce; combines search, recommendations, and browse with ML-driven ranking.
Database-Embedded Search (SQL)
Search run inside an existing database rather than a dedicated engine — increasingly viable for small-to-medium workloads, eliminating a separate search cluster.
- PostgreSQL — native Full-Text Search (
tsvector, GIN,ts_rank), plus pgvector for vector search and ParadeDB for BM25; supports Hybrid Search + RRF in SQL. See Search using PostgreSQL.
Key Trade-offs When Choosing a Platform
| Dimension | Open Source / Self-Hosted | SaaS / Proprietary |
|---|---|---|
| Control | Full (data, config, infra) | Limited |
| Ops burden | High | Low |
| Cost at scale | Lower (infra only) | Higher (per-query or per-record pricing) |
| Vendor lock-in | None | High |
| Time to first result | Slower | Fast |
Related Notes
- E-commerce Search — platform choice is a recurring theme in ecommerce search case studies
- Enterprise Search — enterprise teams frequently operate Elasticsearch or Vespa at scale
- Hybrid Search Blueprint Series Semantic Boosting — hybrid retrieval patterns that span Lucene and vector platforms
- Vinted - Migrating Search from Elasticsearch to Vespa · Kleinanzeigen - Vespa Migration for Homepage Feed — real ES→Vespa migration case studies
- Conversational and Agentic Search — agentic workloads impose distinct retrieval requirements; Hornet is built specifically for this use case
- Search using PostgreSQL — running modern search inside Postgres (FTS + pgvector + ParadeDB + RRF)
- Extreme Search Systems — the scale dimensions that decide which platform survives
- turbopuffer Search DB — object-storage-native engine; a different set of trades from both self-hosted clusters and managed vector DBs