Vald is an open-source, highly scalable distributed vector search engine built for the embedding era. It has 1,734 stars on GitHub and is under active development in 2026. Vald powers similarity search over billions of feature vectors, the retrieval backbone behind recommendation, image search, and semantic search systems.
Vald is designed around a cloud-native architecture and runs on Kubernetes. At its core it uses NGT (Neighborhood Graph and Tree), one of the fastest approximate nearest neighbor algorithms, to search high-dimensional dense vectors at production scale. The system handles automatic vector indexing and index backup, and it scales horizontally by adding nodes, so the same deployment pattern works from a prototype index to billions of embeddings. Written in Go, Vald exposes a gRPC-based API and ships with Helm charts for repeatable installs. A Vald cluster is made of gateway, agent, and manager components that can be scaled independently, and indexes can be backed up for disaster recovery. The project is listed in the CNCF landscape and publishes its Helm chart through Artifact Hub.
Vald is free and open source under the Apache-2.0 license. You run it on your own infrastructure, so costs are limited to your Kubernetes environment.
Vald is for platform and search engineers building large-scale similarity search: image and video retrieval, recommendation systems, and embedding-based semantic search that has outgrown single-node libraries. Its operational model suits teams that need predictable latency at very large index sizes.