eg
Vespa logo

Vespa

visit →

AI-powered search engine and vector database

paid· from $0.10/mo· for enterprise· verified 9 days ago

Vespa runs vector, lexical, and structured search in a single engine, with machine-learned ranking applied at query time across billions of items. Ecommerce platform and search teams use it to power product discovery, personalized recommendations, and semantic retrieval at catalog scales where Elasticsearch or standalone vector databases hit latency ceilings. The buyer is typically a CTO or search engineering lead replacing a stitched-together stack with one system handling retrieval, ranking, and serving.

> pick this if

Pick this if you're a search or platform engineering leader running catalogs in the tens of millions to billions of items and you want vector, lexical, structured filters, and ML ranking executed in one engine rather than stitched across Elasticsearch, a vector DB, and a re-ranker.

> look elsewhere if

Look elsewhere if you're a mid-market merchant under ~$50M GMV without a dedicated search engineering team — Algolia, Typesense, or Shopify-native search will ship faster with a fraction of the operational burden.

> Vespa is used by

  • Metal AI
  • Clarm
  • Perplexity
  • Spotify
  • Elicit
  • Yahoo
  • Onyx
  • Mimeta – Civsy
  • Qwant
  • Vinted
  • RavenPack Bigdata.com

> Vespa is built for

  • platform-agnostic

> what it does for ecommerce

  • Combines vector, lexical, and structured filters in one query pass
  • Runs ML ranking models inline during retrieval, not post-hoc
  • Co-locates embeddings with metadata to avoid cross-system joins
  • Scales to billions of documents with sub-100ms tail latency
  • Available as managed Vespa Cloud or self-hosted open source

> how you'd use it

  • Marketplace or large catalog retailer, $200M+ GMV, 8–20 person search/platform engineering team
    Replacing a stack of Elasticsearch for lexical plus a separate vector DB (Pinecone/Weaviate) for semantic retrieval, where re-ranking happens in a downstream service and tail latency exceeds 300ms on 50M+ SKUs
    → Single-pass hybrid retrieval with inline learned ranking, sub-100ms p99 on billions of documents, and one system to operate instead of three
  • DTC brand group or retailer with long-tail catalog, $50M–$500M GMV, dedicated ML/search team of 4–10
    Serving personalized product discovery where embeddings, business rules (margin, stock, boost flags), and user features all need to influence ranking at query time
    → ML ranking models (GBDT, neural) execute during retrieval with user and item features co-located, lifting conversion on search and PLPs without a separate re-ranking microservice
  • Commerce platform or headless SaaS vendor serving multi-tenant merchants, enterprise scale
    Building semantic site search and recommendations as a product feature across thousands of merchant catalogs with strict latency SLAs
    → Vespa Cloud handles tenant isolation, real-time indexing, and mixed vector/lexical/structured queries; engineering avoids building a bespoke retrieval platform

> Vespa use cases

> Vespa key features

  • Real-time machine-learned model inference
  • Supports vector search and lexical search
  • High scalability and performance
  • Open-source cloud service
  • Co-location of vectors and metadata

> Vespa pricing

verified 9 days ago
  • Startup

    Plan for testing and getting your business started with restricted features

    • Restricted features
    • Community support only, no SLA
    • Runs on shared resources
    • No SSO, no autoscaling
    • No redundancy by default
    • No CI/CD pipeline
    • No testing before Vespa upgrades
    • Dev zones only
    • Approx. 15 mins/day downtime during upgrades

    Low fixed unit cost

  • Basic

    Suitable for applications that don't need 24/7 operational support

    • Prices go down with volume
    • Production support response time: Next business day
    • Deployment support response time: Next business day
    • Other support response time: Next 2 business days
  • Commercial

    Suitable for production applications with 24/7 operational support

    • Prices go down with volume
    • Production support response time: 1 hour 24/7
    • Deployment support response time: Next business day
    • Other support response time: Next 2 business days
  • Enterprise

    Suitable for enterprises with 24/7 deployment support

    • Prices go down with volume
    • Production support response time: 15 minutes 24/7
    • Deployment support response time: 1 hour 24/7
    • Other support response time: Next business day
    • Single sign-on (SSO)
    • Named support representative
    • Tune-up program participation
    • Dedicated Slack channel
    • On-site visits

    Minimum monthly spend of $20,000

  • Self Managed

    Self Managed Vespa deployment including support

    • Unlimited support cases

    Contact Sales

Prices vary based on resources allocated and plans chosen. Each support level sets a unit price for machine resources.

> compliance & trust

  • GDPR

> how vespa compares

bidirectional editorial

> alternatives to Vespa in our index

by shared use-case

> Vespa pairs well with