Vespa

What is Vespa?

Vespa is a search and vector database platform for applications that need low-latency computation over large datasets, storing and indexing structured, text and vector data and running ranking and model inference at query time.

Key features of Vespa

• Hybrid search: search text, structured data and vector spaces with approximate nearest neighbor, combined in boolean conditions.
• Ranking: score matches with ranking expressions or ONNX models, with an optional second phase for top candidates.
• Grouping: aggregate results by field values for navigation and analytics.
• Real-time writes: add, replace and remove documents and update individual fields.
• Elastic operation: clusters rebalance data when nodes change and remove failed nodes.
• Visiting: stream matching content out for backup or reprocessing.

How Vespa works

A Vespa application is defined as an application package and deployed to stateless Java container clusters that handle processing and content clusters that store data and run queries. Admin and config clusters manage the system, and the package can be redeployed to change behavior. Vespa Cloud offers a managed option, with a price calculator for estimating cost.

Who uses Vespa?

Vespa is used by engineering teams building search, recommendation and retrieval-augmented generation applications. Its documentation points to e-commerce modules, embedding and LLM integration and a migration guide from Elasticsearch.

Vespa pricing

Vespa does not list prices on its homepage; the Vespa Cloud site points to a price calculator for estimating cost. Self-managed installation documentation is also published for running it yourself.

Vespa integrations

Vespa supports ONNX models for ranking, retrieval-augmented generation and LLM integration, and application components can serve custom network endpoints. Its documentation links to a GitHub organization and an Elasticsearch migration guide.

Vespa alternatives

Weaviate, Qdrant and Milvus are vector databases focused on embedding retrieval. Marqo is a vector search engine with built-in embedding generation. Haystack is a framework for building search and RAG pipelines rather than a database. Vespa combines text, vector and structured search with ranking in one engine.