# Azure OpenAI Vector Store

> Vector stores under an Azure OpenAI account bill per GB per day beyond the 1 GB included free, and the daily charge continues long after the assistant that used them is abandoned. ZopNight discovers each store with size and expiry-policy detail and flags stores that have no expiration policy set.

Source: https://zop.dev/integrations/azure/azure-openai-vector-store
Updated: 2026-08-19

---

Vector stores hold embedded file content for the Assistants file-search capability, billed per GB per day after the free tier. Stores loaded for abandoned assistants keep accruing daily storage charges.

## A storage meter that ticks by the day

A vector store bills on stored gigabytes per day, with 1 GB included before charges begin. The unusual unit is the point: most storage on Azure meters per month, so a vector store's cost accumulates in daily increments that look negligible individually and compound quietly. Nothing about the meter depends on use. A store queried thousands of times a day and a store no assistant has touched since spring bill identically for the same bytes. The only inputs are size and existence, which makes lifecycle discipline the entire cost model for this type.

## Expiration policy is the whole game

Vector stores support an expiry policy that deletes them after a period of inactivity, and that single setting separates hygienic accounts from leaky ones. A store created with an expiration ends itself when its assistant stops calling; a store created without one persists until a human remembers it. ZopNight's data-plane enumeration records each store's size and expiry-policy detail and specifically flags stores with no expiration policy, the population where every long-lived leak on this type begins. Vector stores are discovery-only: there is nothing to stop or schedule, so the flag plus deletion is the remediation path.

## How embedded files outlive their assistants

The common sequence: a team prototypes a file-search assistant, uploads a document corpus, embeds it into a store, and moves on. The assistant is deleted or forgotten; the store is not, because it lives one level removed from the thing the team was iterating on. Multiply by several experiments per quarter and an account accumulates a shelf of orphaned embeddings, each billing daily. Re-embedding a corpus is cheap and reproducible, so keeping stale stores around as insurance rarely pays.

## Finding stores before they age

Azure AI Foundry portal → the account's Vector stores view lists each store with its size and expiration setting; review anything sizable that shows no expiry.