Skip to main content

Project Buckets

Two dedicated cloud storage buckets are automatically provisioned for each project linked to a engine:

BucketName formatPurpose
Inputs{project-namespace}-inputsUpload datasets and artifacts before experiments run
Outputs{project-namespace}-outputsExperiment results and model checkpoints

The {project-namespace} is derived from the project name: lowercased, non-alphanumeric characters removed, truncated to 12 characters, then a 16-character random hex suffix appended (e.g. my-project-abc1234567890ab12).

The {project-namespace} can be obtained from in the Experiments view page by clicking on the Project Details button The name of the buckets can be fetched from the Datasets tab in the Experiments view page.

Cloud provider

Buckets are created on the same cloud provider as the project engine:

Engine typeStorage backendS3 endpointProtocol
GCP (GKE)Google Cloud Storagehttps://storage.googleapis.comS3-compatible (HMAC)
AWS (EKS)Amazon S3AWS regional endpointNative S3 (STS)
Azure (AKS)Azure Blob Storagehttps://{account}.blob.core.windows.netAzure Blob API
On-premisesGCP by default, customizable on request--

Lifecycle and data retention

The inputs and outputs buckets have no automatic expiry policy. Data stored in them persists indefinitely until manually deleted.

Buckets are not deleted when a project is deleted

When a project is deleted in AIchor, the cloud storage buckets are not automatically removed from the cloud provider account. They remain and continue to accrue storage costs. After deleting a project, the bucket contents and the buckets themselves must be cleaned up manually.

File and folder naming

Some special characters cause issues across third-party tools and storage providers. Use only the following characters in file names, folder names, and paths:

  • 0-9
  • a-z
  • A-Z
  • Hyphen (-)
  • Underscore (_)
  • Period (.)

A single file or folder name must therefore match:

^[A-Za-z0-9._-]+$

A full path adds the / separator between names:

^[A-Za-z0-9._/-]+$

Storage relared environment variables in experiments

AIchor injects a number of environment variables related to storage into every experiment pod. The full list of predefined variables, with example values, is on the Environment Variables page.