Skip to Content
We are live but in Staging 🎉
Core Concepts

Core Concepts

The two planes

VBase has a control plane and a data plane, and you use different tools for each:

PlaneWhat it doesHow you use it
Control planeProvision + access: allocate a database, fetch its connection details, list, deleteVBase API / dodil vbase db … (CLI, API)
Data planeEverything inside a database: collections, schemas, indexes, insert, searchThe Milvus SDK, pointed at the database (Connecting)

Databases

A database is what you allocate from VBase — your isolated vector store, identified by a service_id and reached at an endpoint + port with a db_name. It’s powered by Milvus 2.6 and is the unit of provisioning, access, and isolation.

Allocation is serverless: a database is placed on managed, multi-tenant Milvus infrastructure on demand — there’s nothing for you to size or run. Dedicated clusters (a database backed by isolated capacity) are a preview feature coming later.

Allocation is asynchronous — a new database reports CREATING, then RUNNING. Wait for RUNNING before connecting (the Quickstart shows the poll).

What VBase manages vs what you manage

VBase manages:

  • The Milvus infrastructure your database runs on.
  • The Milvus users and roles behind your database — you authenticate with your IAM token and VBase maps it to the right Milvus credentials. You don’t (and can’t) run Milvus user/role admin commands; that’s owned by VBase.

You manage (through the Milvus SDK):

  • Collections and their schemas.
  • Indexes (HNSW, IVF, …) and metrics (L2, IP, COSINE).
  • Data — insert, upsert, delete.
  • Search — vector, filtered, and hybrid + BM25.

Powered by Milvus 2.6

VBase exposes the native Milvus 2.6 API — the data plane is Milvus, reached through a transparent layer that adds IAM auth and usage metering. Because it’s the real Milvus protocol, the standard SDKs and the full feature set apply, and the Milvus documentation  is your reference for all collection/index/search operations.

VBase and K3 Vector

K3 ’s Vector pillar is built on VBase: it allocates a VBase database for you and wraps it with curated defaults (fixed shard count, strong consistency, a managed service account, a limited operation set). Going direct to VBase trades that convenience for the full native Milvus surface and control over your own database lifecycle, schemas, indexes, and naming.


See also