Vald
activeby LY Corporation (vdaas) · since 2020
Kubernetes-native distributed vector search engine built on Yahoo Japan's NGT.
Overview
Vald is a distributed approximate nearest neighbour search engine created at Yahoo Japan (now LY Corporation) and designed to run only on Kubernetes. It uses the NGT graph index, shards vectors across agent pods, rebuilds indexes asynchronously without blocking queries and backs index state up to persistent volumes or object storage. Filtering and pre/post-processing are handled by pluggable gRPC 'filter' components. It is solid infrastructure for teams already deep in Kubernetes, but it has a much smaller community than Milvus or Qdrant and no metadata-rich query layer.
Strengths & trade-offs
Strengths
- +Asynchronous auto-indexing keeps queries serving during rebuilds
- +Horizontal scaling and self-healing via Kubernetes
- +NGT index offers strong speed/recall trade-offs
- +Pluggable ingress/egress filters for custom preprocessing
Trade-offs
- −Kubernetes is mandatory
- −Minimal metadata filtering and no hybrid/full-text search
- −Small community and ecosystem integrations
- −Slow release cadence in 2024-25
Great for
- +Large-scale image or embedding similarity search on Kubernetes
- +Recommendation candidate retrieval
- +Internal ANN services
Think twice for
- −Non-Kubernetes environments
- −RAG needing rich metadata filters
- −Quick prototypes
Capabilities
Native or official first-party support · 2 of 25 tracked
Scale & consistency
- Comfortable data size
- Large · ≤ 1 PB
- Horizontal scaling
- Native
- Multi-region
- No
- Transactions
- None
- Consistency
- Eventual
- Schema
- Schemaless
- Architecture
- Distributed, Client server
- Query languages
- Vald gRPC API
Agents shard and replicate the index across Kubernetes pods; LB gateway fans queries out and merges results.
Deployment
Ecosystem
Alternatives to Vald
Cloud-native, distributed open-source vector database built for billions of vectors.
vald vs milvus →Rust vector search engine with rich payload filtering, quantization and a managed cloud.
vald vs qdrant →Open-source AI-native vector database with hybrid search, modules for embeddings and multi-tenancy.
vald vs weaviate →Big-data serving engine for search, recommendation and RAG with tensors and ML ranking.
vald vs vespa →Meta's C++/Python library for fast dense-vector similarity search and clustering, CPU and GPU.
vald vs faiss →Google Cloud's managed ANN service built on ScaNN, now part of Gemini Enterprise Agent Platform.
vald vs vertex-ai-vector-search →Vald FAQ
What is Vald?
Vald is a vector database by LY Corporation (vdaas), first released in 2020. Vald is a distributed approximate nearest neighbour search engine created at Yahoo Japan (now LY Corporation) and designed to run only on Kubernetes. It uses the NGT graph index, shards vectors across agent pods, rebuilds indexes asynchronously without blocking queries and backs index state up to persistent volumes or object storage. Filtering and pre/post-processing are handled by pluggable gRPC 'filter' components. It is solid infrastructure for teams already deep in Kubernetes, but it has a much smaller community than Milvus or Qdrant and no metadata-rich query layer.
Is Vald open source?
Yes. Vald is open source. License: Apache License 2.0.
Does Vald support ACID transactions?
No. Vald does not provide ACID transactions; its consistency model is eventual.
Does Vald scale horizontally?
Vald scales horizontally out of the box, with built-in sharding or partitioning. It is comfortable at large scale (≤ 1 PB).
How can I run Vald?
You can self-host it.
What is Vald best for?
Vald is a strong fit for large-scale image or embedding similarity search on Kubernetes, recommendation candidate retrieval and internal ANN services. Think twice for non-Kubernetes environments.
What are the best alternatives to Vald?
Common alternatives to Vald include Milvus, Qdrant, Weaviate, Vespa and Faiss. The right choice depends on your workload, scale and hosting; WhichDB's smart match ranks them against your requirements.
How popular is Vald?
Vald ranks #382 of 844 databases on the WhichDB Score (5.6/100), and #24 among vector databases. The score combines public adoption signals such as package downloads, GitHub activity, job posts and developer discussion.