Vald
Vald is a Kubernetes-oriented distributed ANN engine, not an unaudited "fastest NGT" census. Rechecked 2026-08-18: github.com/vdaas/vald had 1,716 stars, license Apache-2.0, homepage vald.vdaas.org, description A Highly Scalable Distributed Vector Search Engine. vald.vdaas.org still says it uses NGT, automatic indexing / backup, and horizontal scale, and showed docs version v1.7.16. The old vald.dev host failed TLS from this environment; treat vald.vdaas.org as the live site. There is no public dollar seat. Drop leftover copy that never printed stars, license, or the current domain.
Compare Qdrant if you wanted a larger OSS engine plus Cloud Free / usage-based Standard, Milvus if you wanted another OSS vector DB (cloud is Zilliz), and Faiss if you wanted an in-process MIT library instead of a Kubernetes service.
Key Features
- OSS, self-host on Kubernetes: README requires Kubernetes 1.19+ and AVX2 for the NGT agent. Helm chart is on vald.vdaas.org/charts. That is the product, not a hosted $20 row.
- NGT is the ANN backend: First-party pages still name NGT. "Fastest ANN" is their marketing; we did not re-run ann-benchmarks.com.
- No Cloud seat on the homepage: Docs, Slack, and Helm are public. We did not invent a Vald SaaS dollar.
- Website moved: Point new copy at vald.vdaas.org. Recheck vald.dev only if TLS comes back.
Use Cases
- People who will run Helm on Kubernetes and pay their own cluster.
- Teams comparing 1,716 stars against Qdrant 34,018 or Milvus 45,663.
- People who already standardized on NGT and want a distributed wrapper.
Limitation: We did not re-measure billion-vector claims. Treat that as first-party marketing, not a census.
Pricing
GitHub and vald.vdaas.org on 2026-08-18.
| Piece | Price | Notes from first-party pages |
|---|---|---|
| Vald OSS | $0 | Apache-2.0. 1,716 stars. Self-host on Kubernetes. |
| Your cluster | Your infra | AVX2 agents, Helm. Not a Vald SKU. |
| Managed vector Clouds | Recheck that product | Qdrant Cloud / Zilliz / Pinecone are other products. |
If a leftover blog still points only at vald.dev with no license, treat this table as the source.
Getting Started
- Open vald.vdaas.org or github.com/vdaas/vald.
- Recheck the Helm get-started page before you assume a Cloud console.
- Recheck Qdrant if you wanted a printed Cloud Free node.
- Do not paste "fastest ANN" as a measured result.
Frequently Asked Questions
Is it free?
Yes. Apache-2.0. You pay the Kubernetes cluster.
Is vald.dev still official?
The GitHub homepage and the site we opened are vald.vdaas.org. vald.dev failed TLS here.
Same as Qdrant?
No. Qdrant has a much larger star count and a public Cloud matrix. Vald is a smaller NGT-on-Kubernetes engine.
Alternatives
- Qdrant: OSS plus Cloud Free / usage-based Standard.
- Milvus: Another open vector DB; managed Cloud is Zilliz.
- Faiss: MIT library, not a Kubernetes service.
- Vespa: Search + ranking serving engine that also does vectors.
Tips
- Quote 1,716 stars, Apache-2.0, and vald.vdaas.org.
- Call NGT "fastest" only as their claim.
- Recheck Qdrant/Milvus if you wanted a Cloud seat.
Conclusion
Vald is an Apache-2.0 Kubernetes ANN engine with 1,716 stars on vald.vdaas.org, not an unaudited fastest-NGT census. Start at vald.vdaas.org, then decide whether Qdrant already covers the Cloud table you need.
Comments
No comments yet. Be the first to comment!
Related Tools
Vespa
vespa.ai
OSS search+vector engine. Rechecked: Apache-2.0, 7,052 stars. Cloud has a free trial; unit table exists. Not an unpriced toolkit.
PGVector
github.com/pgvector/pgvector
Postgres vector extension. Rechecked: PostgreSQL license, 22,651 stars. Not a Faiss wrapper. No hosted dollar seat.
Elasticsearch
www.elastic.co/elasticsearch
Distributed search engine. Rechecked: 77,791 stars, AGPL/SSPL/Elastic License; Cloud trial + VCU units. Not leftover 8.0-only.
Related Insights

Grok Bot and Hermes Bot: one person finally gets a think tank and a secretariat
Grok Bot now ships with Cursor Pro+. Hermes Bot runs on a VPS. They are not smarter chat boxes. The think tank advises, the secretariat executes, and you still make the call.
Hook OpenCode Go into Codex on Windows. Do Not Open a Second Toolkit.
A ChatGPT-signed Codex desktop app still shows mostly GPT in the picker. On Windows, enable only OpenCode Go and the Grok, GLM, Kimi, DeepSeek, and MiniMax models you already pay for appear in the same selector. Keys stay local. Native GPT stays put.
What Locks Codex Is the Picker, Not the Models
You already pay for OpenCode Go, Grok, and Z.ai, but the Codex picker still shows mostly GPT. The community project codex-router does not teach another install ritual. It puts subscriptions you already bought back into the selector, keeps keys on the machine, and leaves native GPT alone.