---
title: "Vector database concepts"
description: "The terms behind a Qdrant vector database in TIR, such as points, collections, shards and replicas, and how many vectors a plan holds."
content_type: "concept"
product: "vector-database"
platform: "TIR"
url: "/tir/vector-database/concepts/vector-database-concepts/"
---

# Vector database concepts

A vector database stores vectors and finds the ones most similar to a query. A vector is a list of
numbers that a machine learning model produces for a piece of text, an image or other data. These
terms appear when you create and use a database in TIR.

| Term               | What it means                                                                    |
| ------------------ | -------------------------------------------------------------------------------- |
| Point              | A record made of a vector and an optional payload.                               |
| Vector             | The numbers that represent your data in many dimensions.                         |
| Collection         | A container for related vectors, like a table in a database.                     |
| Node               | One server in the database's cluster.                                            |
| Shard              | An independent partition of the points in a collection, spread across nodes.     |
| Replication factor | How many copies of each shard the cluster keeps, for protection if a node fails. |
| Index              | The vector and payload index that makes filtered similarity search fast.         |

## Nodes, shards and copies

A cluster of three or more nodes can keep running if one node fails. A higher replication factor makes
the database more resilient. In the portal, a database has 3 to 10 nodes and its default collection
has a replication factor of at least 3. Setting the shard number to a multiple of the node count
spreads the data evenly. When you add nodes, they start empty; see
[Use new nodes after you scale up](/tir/vector-database/how-to/use-new-nodes/).

## How many vectors a plan holds

**Create Database Cluster** shows an estimate of how many million vectors your plan holds. It uses the
vector dimension (128, 256, 512, 1024 or 4096) and the replication factor you pick in the estimate,
and shows the RAM, vCPUs and node count of the cluster. Use it to choose a plan before you launch. See
[Create a vector database](/tir/vector-database/how-to/create-a-vector-database/).

## Good habits

- Take a [snapshot](/tir/vector-database/how-to/take-a-snapshot/) before you scale up or change a lot of
  data.
- Check **Total pending operations** after scaling; see
  [Monitor a vector database](/tir/vector-database/how-to/monitor-a-vector-database/).
- Give client applications that only query data the **Read-only API Key**, and keep every API key out
  of client-side code.