docs/en/platform/integrations/azure-blob-storage.md
The Azure Blob Storage integration connects your storage account containers to Ultralytics Platform. Your images stay in your containers — Platform indexes them in place, so you can browse, annotate, and train YOLO models without uploading a copy.
!!! note "Pro feature"
Azure Blob Storage datasets require a [Pro or Enterprise plan](../account/billing.md#plans). Free workspaces see the integration and are prompted to upgrade when connecting. Existing Azure Blob Storage datasets stay fully accessible if a subscription ends — only new connections and imports require Pro.
Platform only ever reads from your storage — it never writes, modifies, or deletes your blobs. The current integration requires an account access-key connection string, which grants account-wide privileges even though Platform uses only list and read operations:
!!! note "Public Azure cloud only"
Connections use the standard `blob.core.windows.net` endpoint. Sovereign clouds (Azure China, Azure Government) and custom blob endpoints are not supported.
Reconnecting the same storage account later adds new containers to the existing integration. A saved credential is only replaced once its replacement can still read every container you've already connected.
!!! note "Credential security"
An account key can authorize write and delete operations or create SAS tokens if it is exposed outside Platform. Credentials are encrypted at rest with AES-256-GCM, are never returned to the browser, and never enter training job payloads. Use a dedicated storage account where practical. To revoke access, rotate the storage account access keys in Azure.
Platform lists the folder once and indexes what it finds:
.jpg, .jpeg, .png, .webp, and .avif blobs are indexed with dimensions read from bounded header requests. Source pixels are never copied out of your container..txt sidecars are parsed into Platform annotations, matched by the standard images/ → labels/ layout or as same-folder siblings.data.yaml/data.yml provides class names, task type, and pose keypoint shape, exactly like an archive upload.train, val, and test folder names in the blob path assign splits automatically.The dataset then behaves like any other: browse and annotate it, set it public or private, share it with your team, and train on it through managed training. Originals are streamed on demand, and indexed images do not consume your Platform storage quota.
!!! note "Limits"
A single import indexes up to 50,000 blobs, and label or YAML files up to 1 MB each. Larger containers should be split across multiple datasets.
!!! warning "Keep indexed blobs immutable"
Every indexed image is pinned to its blob ETag, and Platform fails closed if a blob changes underneath it. Add new blobs instead of overwriting existing ones.
If an import fails — an empty folder, a typo in the path, or revoked permissions — the dataset shows the error on its page. Editors can click Retry import to restart it with the stored container and folder, or create a new dataset pointing at the corrected path.
Managed training works through the normal training flow. Workers download the pinned originals into temporary job storage for the run and remove them with job cleanup — your Azure credentials never reach compute.
Azure-backed datasets currently exclude features that require Platform-owned copies of your images: auto-annotation, clustering analysis, dataset cloning, and immutable version snapshots.
Deleting an Azure-backed dataset, or individual images from it, removes Platform's references only — your blobs are never touched.
Also see the Google Cloud Storage and Amazon S3 integrations.