Vector Database Overview

Vector Database (VBase) is the vector database layer of DODIL. It exists to make storing, searching, and operating on vectors reliable in production, not just in demos built on top of open-source Milvus.

Most teams can get a vector search prototype working quickly. The hard part starts when you need to run it as a real service: many tenants, many datasets, high throughput, predictable latency, access control, cost tracking, and safe operations over time.

VBase solves that gap.

Why VBase exists

Vector search becomes “infrastructure” the moment it powers real user experiences:

semantic search across documents
RAG over knowledge bases
recommendation and similarity matching
deduplication and near-duplicate detection
retrieval over multimodal embeddings (text, images, audio, code)

At that point, you need more than “a database that supports vectors.” You need a managed vector service with consistent behavior, operational safety, and clear control over performance and cost.

VBase exists to provide that production foundation.

What problem VBase solves

1) Operating vectors at scale

Handling millions to billions of vectors requires careful control over:

indexing and compaction behavior
memory and storage usage
query throughput and tail latency
background maintenance and upgrades

VBase provides the service layer that makes these concerns manageable.

2) Multi-tenant isolation and governance

In real systems, many organizations and teams share the platform. VBase is designed for:

tenant isolation (data + performance boundaries)
consistent lifecycle management (create, resize, move, delete)
governance and access patterns that match enterprise needs

3) Predictable performance

VBase focuses on stable production behavior:

durable writes
high-availability options
consistent query behavior under load
operational controls to avoid “works in dev, breaks in prod”

4) A clean developer experience

VBase is built to be consumed by apps and pipelines:

simple APIs for collections, indexes, inserts, and search
compatibility with ingestion pipelines (like VNG)
observability hooks for monitoring and debugging

What VBase provides

Vector storage + search as a service
Collection and index management
Hybrid retrieval patterns (vectors + filters/metadata)
Operational controls for scale, upgrades, and maintenance
Multi-tenant readiness for real platforms
Observability: metrics, logs, tracing-friendly behavior
Cost-aware operation: built for measured usage and planning

VBase is Milvus-backed, but productized as a managed, secure, and scalable service layer.

When to use VBase

Use VBase when you need vector search to be a dependable platform component:

you have multiple apps/teams/tenants
you expect growth in data volume or traffic
you need stable latency and safe operations
you want to avoid building and maintaining vector DB infrastructure yourself

Next steps

Create your first collection
Insert vectors with metadata
Run similarity search and filtered retrieval
Connect VNG to continuously ingest your datasets