Vespa AI features, pricing and GDPR (October 2026)

Vespa is a platform for search, recommendation and retrieval for AI applications. It combines vector search, keyword search, structured filters and machine-learned ranking in one engine, available as open source or as the managed service Vespa Cloud.

Vespa.ai AS is headquartered in Trondheim, Norway, where the technology began as a web search project in the early 2000s. The engine handles text, vectors, tensors and structured data, and the vendor states it scales to billions of items and thousands of queries per second. Vespa Cloud adds managed operations, autoscaling, a continuous deployment pipeline and mutual TLS. On-premises support is available too. The vendor names Spotify, Yahoo, Farfetch, Elicit and Perplexity as users in case studies. Developers describe schemas and ranking in an application package.

• Retrieves context for RAG and AI agentsQueries combine vector similarity, keyword matching and structured filters, then rank passages in phases. The language model receives better context, which the vendor presents as the main lever for answer quality. Engineers tune one ranking pipeline instead of several services.
• Runs ranking models inside the searchVespa evaluates machine-learned models, such as ONNX, XGBoost and LightGBM models or cross-encoders per its documentation, on candidates while the query runs. Relevance teams improve results with trained models without moving data to an external ranking service.
• Recommends and personalizes contentThe system retrieves eligible items and scores them per user with learned models. The vendor lists recommendation, personalization and ad targeting as use cases, so product teams can serve individual results at low latency from the same data that powers search.
• Searches private data per userA streaming search mode skips building indexes for applications where each query touches only a small share of the data, for example personal collections. The vendor states this is much cheaper than indexing for such workloads, and it keeps vector and ranking features available.

• Retrieval, ranking and model inference run in one distributed engine instead of separate search, vector and ranking services.
• Available as open-source Vespa, as Vespa Cloud, or with on-premises support, so teams can move between operating models.
• Pricing is by allocated resources per hour, with support levels from Startup to Enterprise, and no other charges for platform features according to the pricing page.
• Streaming search mode targets personal and private data, where the vendor says it costs far less than indexing.

Headquarters and engineering hub are in Trondheim, Norway, which is outside the markets this directory lists. Vespa Cloud production zones exist in North America, Europe and Asia, so no single focus market is claimed.

Expect an engineering project: schemas, ranking and deployment are defined in code, so the platform suits teams with search or ML engineers. Production use starts at the Basic level. The Startup plan runs on shared resources in development zones without SSO or an SLA, and Enterprise requires a minimum monthly spend of $20,000. The vendor states a SOC 2 attestation without naming the type, so request the report. We found no public data processing agreement in October 2026.

No reviews yet. Be the first to share how your team uses it.

Every review shows its verification level. We read each one before it goes live. How we check reviews