KubeDB

By AppsCode

CONTACT SALES

Vector Database on Kubernetes

Kubernetes-Native RAG and
Vector Database Infrastructure

Run Qdrant, Milvus, Weaviate, and PostgreSQL with pgvector on Kubernetes, next to your models and applications. KubeDB provisions, scales, secures, backs up, and monitors the data layer behind retrieval-augmented generation, semantic search, and AI agents.

Best with:
KubeDB Operator
KubeDB Platform
Explore Vector Databases Schedule Demo

4

Vector Engines

Any Engine

GitOps Support

Automated

Day-2 Operations

24x7

Support

Why Production AI Applications Need Kubernetes-Native Vector Databases

Embeddings, indexes, and metadata are stateful data. Treat them with the same care as any production database.

Vector Data Is Stateful

Embeddings and indexes live on disk. Losing them means re-embedding your corpus, so storage, backup, and upgrades matter.

Your Data Stays Yours

Run the vector store in your own cluster, VPC, data center, or air-gapped site, next to your models and applications.

The Right Engine for Each Workload

Choose Qdrant, Milvus, Weaviate, or PostgreSQL with pgvector, and operate them all the same way.

Day-2 Without Custom Scripts

Scaling, TLS, auth rotation, reconfiguration, and monitoring are Kubernetes resources, not runbooks.

Manage Qdrant, Milvus, Weaviate, and pgvector on Kubernetes

Pick the engine that fits your workload. KubeDB gives each one the same declarative, Kubernetes-native lifecycle.

EngineBest forTopologiesWhat KubeDB automates
QdrantHigh-performance similarity search with payload filteringStandalone, or distributed with Raft consensusTLS, logical backup, volume snapshots, monitoring, scaling, autoscaling, and GitOps
MilvusEmbedding search over very large collections, RAG, and recommendationsStandalone or distributed, with external object storage and KubeDB-managed or external etcdBackup with volume snapshots and PITR, scaling, volume expansion, storage migration, TLS, auth rotation, version updates, autoscaling, and GitOps
WeaviateObjects and vectors together, with semantic and hybrid searchMulti-node clusterBackup and restore, API key auth, TLS, scaling, volume expansion, storage migration, autoscaling, and GitOps
PostgreSQL with pgvectorEmbeddings next to relational data, queried with SQLStandalone or HA cluster, with cross-data-center DRHNSW and IVFFlat indexes, automatic failover, backup with point-in-time recovery, monitoring, and GitOps
Your Infrastructure, Our Software

RAG Infrastructure, Operated for You

Your application and models stay where they are. KubeDB runs and protects the data layer they depend on.

You Provide:

A Kubernetes cluster, in the cloud, on-prem, or air-gapped
Your embedding model and LLM, hosted wherever you choose
A StorageClass for vector data, and object storage for backups
Your RAG service, search API, or agent

KubeDB Provides:

Provisioning of Qdrant, Milvus, Weaviate, or PostgreSQL with pgvector from one manifest
TLS, authentication secrets, and rotation
Scaling and autoscaling as your corpus grows
Backup, monitoring, and version updates
Declarative • Self-hosted • Same operator as your other databases

Everything Your Vector Databases Need in Production

Provisioning to recovery, handled by the same operator you use for your other databases

Declarative Provisioning

Deploy a vector database with one Kubernetes manifest, standalone or distributed.

Qdrant, Milvus, Weaviate, and pgvector
Standalone and distributed topologies
GitOps support for any engine

Scale With Your Corpus

Grow compute and storage through ops requests, or let the autoscaler do it.

Vertical and horizontal scaling
Volume expansion and storage migration
Compute and storage autoscaling

Security and Access Control

Encrypt traffic and control who can reach your embeddings.

TLS, with certificate reconfiguration
Authentication secrets you can rotate
API key authentication for Weaviate

Backup and Recovery

Protect the data you spent compute to create, on every supported engine.

Backup and restore for Qdrant, Milvus, Weaviate, and pgvector
Volume snapshots, logical backups, and PITR where the engine supports them
Any S3-compatible object store

Observability

See how your vector stores behave under real query load.

Prometheus and Grafana monitoring
Metrics for Qdrant, Milvus, and Weaviate
One approach across all your databases

Resilience

Keep retrieval available through node and site failures.

Qdrant distributed mode with Raft consensus
Automatic failover for PostgreSQL with pgvector
Cross-data-center DR for pgvector databases, see BCDR
WHY KUBEDB

What Teams Build with KubeDB and Vector Search

From a first RAG prototype to a regulated production system

Retrieval-Augmented Generation

Ground LLM answers in your own documents with a vector store that runs next to your application.

Semantic Search

Search by meaning across documents, tickets, code, and product catalogs.

Recommendations

Match users and items by embedding similarity, with an engine built for large collections.

AI Agents with Memory

Give agents long-term memory backed by a managed, backed-up vector store.

Private and Regulated AI

Keep embeddings and source data inside your network, including on-prem and air-gapped sites.

Hybrid Relational and Vector

Store embeddings beside transactional data with pgvector, and query both with SQL.

The Process

From Manifest to Production

A vector database is a Kubernetes resource. Declare it, connect your application, and add protection as you grow.

1
Step 1

Install KubeDB

Install the operator on any Kubernetes cluster, including air-gapped ones

2
Step 2

Choose an Engine

Qdrant, Milvus, Weaviate, or PostgreSQL with pgvector, standalone or distributed

3
Step 3

Apply One Manifest

Declare version, topology, storage, and TLS in a custom resource

4
Step 4

Connect Your App

Use the generated service and auth secret from your RAG service or agent

5
Step 5

Load Embeddings

Ingest vectors with your own pipeline and the engine's native client

6
Step 6

Protect and Scale

Add backup, monitoring, and autoscaling as the workload grows

FAQ

AI + KubeDB Frequently Asked Questions

Everything you need to know about running vector databases with KubeDB

KubeDB provides operators for Qdrant, Milvus, and Weaviate, and supports the pgvector extension on PostgreSQL. The documentation lists the supported versions for each engine.
Use pgvector when embeddings belong next to relational data and you want to query both with SQL in one database. Use Qdrant, Milvus, or Weaviate when vector search is its own workload, especially for large or fast-growing collections. KubeDB lets you run both side by side.
Yes. Qdrant supports distributed mode with Raft consensus, Milvus supports standalone and distributed topologies with external object storage and KubeDB-managed or external etcd, and Weaviate runs as a multi-node cluster.
KubeDB provides backup and restore for every vector engine it supports. Qdrant supports logical backup and volume snapshots, Milvus supports base backup, volume snapshots, and point-in-time recovery, and PostgreSQL with pgvector has encrypted, incremental backups with point-in-time recovery. Backups go to any S3-compatible object store.
Deploy PostgreSQL with an extension-enabled PostgresVersion, then run CREATE EXTENSION vector in your database. You can then create vector columns and HNSW or IVFFlat indexes.
No. KubeDB runs the data layer: vector and relational databases. Your embedding model and LLM run wherever you choose, such as your own inference service or a provider API.
Yes. KubeDB is self-hosted, so vector data stays in your cluster, and it supports air-gapped installation for isolated environments. See the air-gapped solution for details.
Use ops requests for vertical and horizontal scaling and volume expansion, or enable the autoscaler for compute and storage. Scaling support varies by engine, so check the engine guide for details.
Because pgvector runs on PostgreSQL, it inherits KubeDB PostgreSQL high availability and cross-data-center disaster recovery, with near-zero RPO and RTO. See the BCDR solution for options.
Yes. KubeDB provides GitOps support for all of its databases, including Qdrant, Milvus, Weaviate, and PostgreSQL with pgvector, so you can keep database definitions in Git. See the GitOps solution for the workflow.

Still have more questions?

Contact our support team
Build AI on Solid Ground

Give Your AI Applications a Production-Grade Data Layer

Run vector search where your models and data already live, and operate it like every other database.

  • Qdrant, Milvus, Weaviate, and pgvector
  • Standalone and distributed deployments
  • Backup, monitoring, and autoscaling
Schedule a Demo Read Docs
4
Vector Engines
Auto
Compute and Storage Scaling
TLS
Encryption in Transit
24x7
Expert Support