Overview
Each Mage project has ametadata.yaml file at the root of the project
directory. The file controls storage locations, feature flags, compute
configuration, and defaults that apply to all pipelines in the project.
overrides section (see
/extensibility/env-config/project).
Sample metadata.yaml
Top-level fields
Project layout. Common options:
standalone (default), main, or sub.Optional cluster type used by the project (e.g.,
k8s, docker). Used by Mage Pro
when running managed workspaces.Unique identifier for the project. Generated when the project is created.
Local path for storing pipeline variables and outputs. Paths are relative to
the project root unless an absolute path is provided. Default to
/home/src/mage_data.Remote path (e.g.,
s3://bucket/prefix) for storing variables in object
storage instead of the local filesystem.How long variables are retained (e.g.,
90d).Default project metadata applied to new workspaces created from this project.
Mage Pro workspaces only.
Default workspace configuration applied across environments (e.g., k8s
defaults when running on Kubernetes). Mage Pro workspaces only.
Allows Mage to collect limited telemetry to improve the product.
Feature flags for the project (e.g.,
command_center, dbt_v2,
automatic_kernel_cleanup). Keys are booleans.Whether new or updated triggers are automatically written to code.
Environment-specific overrides for any top-level field. Mage Pro only. See
/extensibility/env-config/project.Compute and execution
Amazon EMR cluster settings (instance types, security groups, key pair, etc.).
Spark configuration shared across pipelines (e.g.,
spark_master,
executor_env, spark_jars, use_custom_session).Project-level defaults for AWS ECS execution.
Project-level defaults for GCP Cloud Run execution.
Project-level defaults for Azure Container Instances execution.
Kubernetes executor defaults applied to pipelines and blocks.
Limits and concurrency settings at the project level.
Configuration for queueing pipeline runs.
State store configuration used by pipelines.
Observability and safety
Project-level alerting configuration (alert types, Slack/Teams webhooks, etc.).
Configure log destinations and formats for the project.
Default retry behavior applied at the project level.
Global AI-related settings (e.g., model providers).
Retrieval-augmented generation settings shared across pipelines.
Project-level OpenAI API key used by AI features when applicable.
Optional LDAP connection settings for authentication.