Vespa

Vespa is an open-source AI search engine, big-data serving engine, and vector database originally developed inside Yahoo and spun out as Vespa.ai AS. Vespa combines vector search, text search (BM25), structured filtering, and machine-learned ranking — including native tensor inference — into a single distributed serving engine that scales to billions of documents with sub-100ms latency. Vespa Cloud is the fully managed commercial offering operated by the Vespa.ai team across AWS and GCP, with Startup, Basic, Commercial, and Enterprise plans plus a Self-Managed option for customers running the open-source engine on their own infrastructure. Vespa is widely used at Spotify, Perplexity, Yahoo, Farfetch, and Elicit for search, recommendation, personalization, and Retrieval-Augmented Generation (RAG).

8 APIs 20 Features

AISearchVector DatabaseBig DataMachine LearningSemantic SearchRetrieval Augmented GenerationOpen SourceTensorRecommendations

APIs

Vespa Query API

The Vespa Query (Search) API executes structured and vector queries against a Vespa application using YQL (Vespa Query Language). It supports text search with BM25, approximate-...

Vespa Document API

The Vespa Document API (/document/v1) provides synchronous REST access to document operations against a Vespa content cluster. It supports Put, Get, Update (partial update with ...

Vespa Deploy API

The Vespa Deploy API (/application/v2) manages application packages on a Vespa configuration server. It supports preparing, activating, and tearing down application packages, se...

Vespa Tenant and Application API

The Vespa Tenant API (/application/v2/tenant) manages tenants and applications hosted on a Vespa configuration server or Vespa Cloud control plane. It exposes operations for cre...

Vespa Config API

The Vespa Config API (/config/v2) lets services in a Vespa application retrieve their configuration from a Vespa configuration server using the config-server / config-proxy prot...

Vespa Cluster Controller API

The Vespa Cluster Controller API (/cluster/v2) exposes runtime state and management endpoints for a Vespa content cluster — including node state queries, maintenance-mode transi...

Vespa State API

The Vespa State API (/state/v1) exposes per-service health, version, and metrics endpoints for any Vespa node — used by orchestration tooling, monitoring agents, and load balanc...

Vespa Metrics API

Vespa exposes a family of metrics endpoints (/metrics/v1, /metrics/v2, /prometheus/v1) that publish Vespa engine and application metrics in JSON or Prometheus exposition format ...

Collections

Vespa Query API

OPEN

Pricing Plans

Vespa Ai Plans Pricing

6 plans

PLANS

Rate Limits

Vespa Ai Rate Limits

6 limits

RATE LIMITS

FinOps

Vespa Ai Finops

FINOPS

Features

Open-source under Apache 2.0

Vector search with HNSW indexes

BM25 text search and hybrid search

Native tensor and ML model inference at serving time

YQL (Vespa Query Language) for structured queries

Multi-phase ranking (match-phase, first-phase, second-phase, global-phase)

Document API with conditional writes, visits, and JSON Lines streaming

Multi-tenant namespaces and document groups

Real-time indexing with sub-100ms query latency

Distributed content clusters with automatic sharding and replication

Streaming search mode for personal/private corpora

Built-in machine learning inference (TensorFlow, ONNX, XGBoost, LightGBM)

Approximate nearest neighbor and exact nearest neighbor operators

Application packages with schemas, services.xml, and rank profiles

Container API for custom searchers, document processors, and handlers

Self-managed (Apache 2.0) or fully managed Vespa Cloud (AWS, GCP)

Vespa Cloud Startup plan from $0.05 / vCPU-hour, $0.005 / GiB-memory-hour

Vespa Cloud Commercial plan with 24/7 1-hour SLA support

Vespa Cloud Enterprise plan with $20k/month minimum and 15-minute SLA

Up to 50% volume discounts and 15% committed-spend discount

Use Cases

Hybrid Search

Combine BM25 text relevance with vector similarity and structured filters in a single query executed by Vespa's multi-phase ranking pipeline.

Retrieval Augmented Generation

Serve grounded context to large language models by indexing documents, chunks, and embeddings in Vespa and retrieving them with hybrid search at sub-100ms latency.

Recommendation and Personalization

Power recommendation systems with machine-learned ranking, real-time feature updates, and tensor inference over user and item embeddings.

Ad Targeting and Real-Time Bidding

Match candidate ads against user context and serve ranked impressions within tight latency budgets using Vespa's distributed serving engine.

E-Commerce Search and Browse

Combine faceted navigation, structured filters, text relevance, and learned ranking for large product catalogs with frequent updates.

Streaming Search for Personal Data

Run "streaming search" mode that scans a user's personal corpus on demand — ideal for mail, messaging, and document search where each user has their own private index.

Integrations

AWS

Google Cloud

Prometheus

Grafana

TensorFlow

ONNX Runtime

XGBoost

LightGBM

Kubernetes

LangChain

LlamaIndex

Haystack

GitHubOrganization

PrometheusExporter

Sources

opencollection: 1.0.0
info:
  name: Vespa Query API
  version: '1.0'
items:
- info:
    name: Query
    type: folder
  items:
  - info:
      name: Execute Query Via GET
      type: http
    http:
      method: GET
      url: http://localhost:8080/search/
      params:
      - name: yql
        value: ''
        type: query
        description: YQL query expression, e.g. `select * from sources * where userQuery()`.
      - name: query
        value: ''
        type: query
        description: Simple user query string passed to the query parser.
      - name: hits
        value: ''
        type: query
        description: Maximum number of hits to return.
      - name: offset
        value: ''
        type: query
        description: Number of hits to skip for pagination.
      - name: ranking.profile
        value: ''
        type: query
        description: Name of the rank profile to use for scoring.
      - name: timeout
        value: ''
        type: query
        description: Query timeout, e.g. `500ms` or `5s`.
      - name: trace.level
        value: ''
        type: query
        description: Trace verbosity level for debugging.
      - name: presentation.format
        value: ''
        type: query
        description: Response format. One of `json`, `cbor`, or `xml`.
    docs: Execute a Vespa query using URL-encoded query-string parameters.
  - info:
      name: Execute Query Via POST
      type: http
    http:
      method: POST
      url: http://localhost:8080/search/
      body:
        type: json
        data: '{}'
    docs: Execute a Vespa query whose parameters are provided as a JSON body. Equivalent to GET but allows longer YQL expressions
      and complex grouping/ranking inputs.
bundled: true

Vespa

APIs

Vespa Query API

Vespa Document API

Vespa Deploy API

Vespa Tenant and Application API

Vespa Config API

Vespa Cluster Controller API

Vespa State API

Vespa Metrics API

Collections

Vespa Query API

Pricing Plans

Vespa Ai Plans Pricing

Rate Limits

Vespa Ai Rate Limits

FinOps

Vespa Ai Finops

Features

Use Cases

Integrations

Semantic Vocabularies

Vespa Ai Context

API Governance Rules

Vespa API Rules

JSON Structure

Vespa Ai Document Structure

Example Payloads

Vespa Ai Document Put Example

Vespa Ai Query Example

Resources

Sources