Resources

Resources are the infrastructure components that Pragmatiks manages: storage buckets, AI models, secrets, agents, and more. You declare resources in YAML, and Pragmatiks handles provisioning, state tracking, and lifecycle management.

Resource Structure

Every resource has a consistent structure:

Field	Description	Example
`provider`	The provider that manages this resource	`agno`
`resource`	The type of resource	`agent`
`name`	A unique identifier you assign	`my-assistant`
`config`	Your settings for this resource	`model_id: claude-sonnet-4-5-20250929`
`tags`	Optional metadata labels	`["env:prod", "team:ml"]`
`depends_on`	Explicit dependencies (optional)	Other resources
`owner_references`	Owner resources for cascade deletion (optional)	Parent resources
`outputs`	What it exposes to other resources (set by provider)	URLs, IDs, secrets

Configuration Example

Here’s an Agno AI model resource:

provider: agno
resource: anthropic-model
name: claude
config:
  model_id: claude-sonnet-4-5-20250929
  api_key:
    provider: pragma
    resource: secret
    name: anthropic-key
    field: outputs.api_key
tags:
  - env:production
  - team:ai

After provisioning, the resource exposes outputs that other resources can reference:

# After provisioning, outputs become available:
# outputs:
#   model_name: "claude-sonnet-4-5-20250929"
#   provider_type: "anthropic"

Resource Lifecycle

Resources move through these states as pragma manages them:

State	Description
`draft`	Created but not queued for processing (use `--draft` flag)
`pending`	Queued for provider to handle. If dependencies aren’t ready, the resource waits in this state until they are.
`processing`	Currently being handled by the provider
`ready`	Successfully provisioned and operational
`failed`	Processing failed, error recorded

A resource in pending state with unready dependencies has no resolved_config — this distinguishes “waiting for dependencies” from “submitted to provider.” Once all dependencies are ready, Pragmatiks resolves the config and submits the resource for processing.

Config vs Resolved Config

Each resource has two config representations:

config — Your original declaration, including FieldReferences and Dependency markers as-is
resolved_config — The fully resolved version where all references have been replaced with actual values from dependency outputs

The resolved_config is what gets sent to the provider. You can inspect both to debug dependency resolution:

pragma resources describe agno/agent my-agent -o json

Idempotent Apply

Re-applying a resource with identical configuration is a no-op. If the resource is already ready and the normalized config matches, Pragmatiks returns the existing resource without re-processing. This makes it safe to run pragma resources apply repeatedly in scripts and CI/CD pipelines.

Deletion Behavior

When you delete a resource, Pragmatiks handles its relationships:

Soft Detach (Dependencies)

Resources that depend on the deleted resource are detached, not cascade-deleted. The dependency reference is removed from their config, and they are sent back to their provider for update. The provider decides whether the resource can continue without the deleted dependency — if it can’t, the resource transitions to failed.

# Before delete:
secret (READY) ← model (READY) ← agent (READY)

# After deleting secret:
model (PENDING → re-processed without secret) ← agent (waits for model)

Cascade Delete (Owner References)

Resources with owner references follow cascade deletion. When the owner is deleted, all owned resources are deleted too. This is used for subresource patterns where child resources have no meaning without their parent.

# Team owns its agents — deleting the team deletes the agents
provider: agno
resource: agent
name: team-member
config:
  instructions: "..."
owner_references:
  - provider: agno
    resource: team
    name: my-team

If a resource has multiple owners, it’s only deleted when the last owner is removed.

Resource Tags

Tags are key-value labels for organizing and filtering resources:

provider: agno
resource: agent
name: my-agent
config:
  instructions: "..."
tags:
  - env:production
  - team:ai
  - cost-center:123

Filter resources by tag:

pragma resources list --tag env:production

Manage tags on existing resources:

pragma resources tags add agno/agent my-agent --tag env:staging
pragma resources tags remove agno/agent my-agent --tag env:production

Applying Resources

Create and provision resources immediately:

pragma resources apply warehouse.yaml

Create resources in draft state (skip provisioning):

pragma resources apply --draft warehouse.yaml

Getting started

Concepts

Guides

Troubleshooting

Resource Structure

Configuration Example

Resource Lifecycle

Config vs Resolved Config

Idempotent Apply

Deletion Behavior

Soft Detach (Dependencies)

Cascade Delete (Owner References)

Resource Tags

Applying Resources

Next Steps

Reactive Dependencies

CLI Resources

​Resource Structure

​Configuration Example

​Resource Lifecycle

​Config vs Resolved Config

​Idempotent Apply

​Deletion Behavior

​Soft Detach (Dependencies)

​Cascade Delete (Owner References)

​Resource Tags

​Applying Resources

​Next Steps

Reactive Dependencies

CLI Resources

Resource Structure

Configuration Example

Resource Lifecycle

Config vs Resolved Config

Idempotent Apply

Deletion Behavior

Soft Detach (Dependencies)

Cascade Delete (Owner References)

Resource Tags

Applying Resources

Next Steps