data.json

{
  "metadata": {
    "name": "Vector Databases for RAG: Deployment-Shape, License, Hybrid-Search and Pricing Comparison 2026",
    "description": "A structured comparison of 11 vector databases for retrieval-augmented generation (RAG) across five dimensions: deployment shape, open-source license, native hybrid search, metadata filtering, and pricing model. Compiled vendor-neutral with conflict disclosure; each cell read from the vendor's own repository, documentation, or pricing page. The thesis: hybrid search and filtering are now table-stakes, so the deciding axis is deployment shape and data gravity, not benchmark.",
    "toolsCompared": 11,
    "deploymentShapes": ["Embedded", "Self-Hosted", "Managed", "Serverless"],
    "dimensions": ["deployment_shape", "open_source_license", "hybrid_search", "metadata_filtering", "pricing_model"],
    "datePublished": "2026-06-17",
    "dateModified": "2026-06-17",
    "license": "CC-BY-4.0",
    "licenseUrl": "https://creativecommons.org/licenses/by/4.0/",
    "canonicalUrl": "https://nesyona.com/research/vector-db-rag-comparison-2026/",
    "methodology": "Each cell read from the vendor's own public repository, documentation, or pricing page as of June 2026. Open-source licenses verified from the project's GitHub repository (Redis confirmed as the Redis 8 tri-license RSALv2/SSPLv1/AGPLv3, not a single permissive license). Hybrid search marked 'yes' only where a single native query path is documented; 'partial' denotes availability through an integration or full-text add-on. Metadata filtering is near-universal across the category. Vendor performance and cost claims are attributed, not asserted; no first-party recall or latency benchmark is encoded here. The four-shape model and the rankings were fixed before any monetization check; no paid placement.",
    "creator": "Vincent Wesley Couey",
    "orcid": "0009-0005-6869-308X",
    "publisher": "Nesyona",
    "definedTerm": "Vector Database Deployment Shapes -- a four-shape reference model (Embedded, Self-Hosted, Managed, Serverless) for choosing a RAG vector store by operational fit and data gravity rather than by recall benchmark, given that hybrid search and metadata filtering are now standard across the category."
  },
  "data": [
    {"tool": "pgvector", "shape": "Extension (in Postgres)", "open_source_license": "PostgreSQL License", "hybrid_search": "partial (Postgres FTS)", "metadata_filtering": "yes (SQL)", "pricing_model": "free (pay your Postgres)", "best_for": "apps already on Postgres, under ~10M vectors"},
    {"tool": "Qdrant", "shape": "Self-Hosted + Managed", "open_source_license": "Apache-2.0", "hybrid_search": "yes (sparse+dense)", "metadata_filtering": "yes (strong, Rust)", "pricing_model": "usage / free tier", "best_for": "fast filtered search at low self-host cost"},
    {"tool": "Weaviate", "shape": "Self-Hosted + Managed", "open_source_license": "BSD-3-Clause", "hybrid_search": "yes (native BM25)", "metadata_filtering": "yes", "pricing_model": "resource-based", "best_for": "native hybrid search and module ecosystem"},
    {"tool": "Milvus / Zilliz", "shape": "Self-Hosted + Managed", "open_source_license": "Apache-2.0", "hybrid_search": "yes (2.4+)", "metadata_filtering": "yes", "pricing_model": "free OSS / usage (Zilliz)", "best_for": "billion-scale distributed workloads"},
    {"tool": "Pinecone", "shape": "Serverless (Managed)", "open_source_license": "proprietary", "hybrid_search": "yes (sparse-dense)", "metadata_filtering": "yes", "pricing_model": "usage (serverless)", "best_for": "zero-ops fully-managed RAG"},
    {"tool": "Chroma", "shape": "Embedded + Cloud", "open_source_license": "Apache-2.0", "hybrid_search": "partial (integration)", "metadata_filtering": "yes", "pricing_model": "free OSS / cloud usage", "best_for": "prototyping and local-first RAG"},
    {"tool": "LanceDB", "shape": "Embedded + Cloud", "open_source_license": "Apache-2.0", "hybrid_search": "partial (FTS)", "metadata_filtering": "yes", "pricing_model": "free OSS / cloud", "best_for": "multimodal and edge / embedded"},
    {"tool": "turbopuffer", "shape": "Serverless (Managed)", "open_source_license": "proprietary", "hybrid_search": "yes (BM25+vector)", "metadata_filtering": "yes", "pricing_model": "usage (object-storage)", "best_for": "storage-heavy, cost-optimized workloads"},
    {"tool": "Vespa", "shape": "Self-Hosted + Managed", "open_source_license": "Apache-2.0", "hybrid_search": "yes (native ranking)", "metadata_filtering": "yes", "pricing_model": "free OSS / resource (Cloud)", "best_for": "complex ML-driven ranking over text+vectors"},
    {"tool": "Marqo", "shape": "Self-Hosted + Cloud", "open_source_license": "Apache-2.0", "hybrid_search": "yes", "metadata_filtering": "yes", "pricing_model": "free OSS / cloud", "best_for": "end-to-end embedding generation + storage"},
    {"tool": "Redis (Query Engine)", "shape": "Self-Hosted + Managed", "open_source_license": "RSALv2 / SSPLv1 / AGPLv3 (Redis 8 tri-license)", "hybrid_search": "yes (vector + text)", "metadata_filtering": "yes", "pricing_model": "free OSS / Redis Cloud", "best_for": "low-latency vectors where Redis is already in the stack"}
  ]
}