Vespa is an open-source platform for real-time vector search, text search, and machine-learned ranking, developed by Yahoo (now Oath/Verizon Media). It combines large-scale serving capabilities with the flexibility of a full-featured search engine, making it suitable for use cases such as recommendation systems, semantic search, personalized feeds, and large-scale retrieval-augmented generation (RAG) pipelines.

Unlike many vector-only databases, Vespa supports hybrid search (combining vector similarity with structured filtering, text relevance, and ML models), enabling complex query logic and custom ranking. It’s optimized for low-latency inference at scale and supports indexing, filtering, and ranking billions of documents in production environments.

Key benefits include:

• Unified support for dense vector search, keyword search, and ML ranking

• Real-time updates, filtering, and aggregation at query time

• Production-ready for large-scale, low-latency applications

What are the main features of Vespa?

Hybrid search engine for vectors, text, and structure

Vespa is designed for flexible, large-scale search across different data modalities.

• Combine dense vector similarity with keyword relevance and structured filters

• Query language supports complex logical conditions, scoring functions, and boosting

• Useful for semantic search, e-commerce, question answering, and personalization

Built-in machine-learned ranking (MLR)

Vespa natively supports ranking using machine learning models, directly during search.

• Deploy linear, tree-based, or ONNX models for scoring

• Apply inference at query time across thousands of candidate results

• Rerank results using custom relevance logic or neural models

Real-time indexing and updates

Vespa provides real-time ingestion and updates without downtime.

• Documents and vectors can be updated individually or in bulk

• Low-latency write path suitable for dynamic content (e.g., news, user behavior)

• Indexes support high availability and consistency

Scalable and distributed architecture

Vespa is built for large-scale deployments, running across multiple nodes with full fault tolerance.

• Horizontally scalable indexing, search, and ranking

• Sharding, replication, and automatic failover included

• Supports billions of documents and large embedding models in production

Advanced filtering and aggregation

Vespa supports complex filtering, grouping, and aggregation during queries.

• Use structured metadata (e.g., user attributes, product categories) in combination with vector similarity

• Compute aggregates, histograms, and top-k results efficiently

• Ideal for personalized ranking and analytics use cases

Why choose Vespa?

• All-in-one retrieval platform: Combine vector, text, and ML-powered search in one system

• Designed for production at scale: Proven in environments with billions of documents and high query volume

• Real-time performance: Ingest, update, and serve with low latency

• Fully open source: No commercial license or usage limits

• Highly configurable: Supports custom query logic, scoring models, and deployment topologies