AMDF - Agnostic Multi-cloud Delivery Framework¶

Transform any Kubernetes CRD into type-safe schemas and developer-friendly blueprints using KCL

Quick Start Examples

Why AMDF?¶

The Problem: CRD Complexity at Scale¶

Modern Kubernetes environments present significant challenges:

• CRD Proliferation: Production clusters typically run dozens of operators (Crossplane, Istio, ACK). Each CRD specification contains hundreds of fields with complex nesting and relationships.

• Configuration Overhead: YAML-based configuration is verbose and lacks type safety, leading to copy-paste patterns and maintenance burden.

• Steep Learning Curve: Teams need deep Kubernetes expertise to effectively use infrastructure resources, creating bottlenecks.

• Delayed Feedback: Configuration errors surface at deployment time rather than during development, slowing iteration cycles.

• Pattern Fragmentation: Without standardization, teams develop inconsistent approaches, making knowledge transfer difficult.

The AMDF Solution¶

AMDF automatically transforms complex CRDs into simple, validated interfaces:

graph LR
    A[Complex CRD] --> B[AMDF] --> C[Type-Safe Schema] --> D[Simple Blueprint]
    style A fill:#f5f5f5
    style D fill:#e8f5e9

Key Benefits¶

How Teams Use AMDF¶

🚀 For Platform Engineers¶

Challenge: Managing numerous CRDs while supporting development teams requires significant effort.

AMDF Value: - Generate once, use everywhere: Create reusable blueprints from any CRD in seconds - Reduce cognitive load: Developers use simple interfaces instead of complex specs - Package & distribute: Share modules via OCI registries (public/private) - Version control: Track changes and roll back when needed

Outcome: Platform team scales efficiently, enabling developer productivity.

🛡️ For DevOps & SREs¶

Challenge: Maintaining configuration consistency and preventing deployment failures across environments.

AMDF Value: - Shift-left validation: Catch errors at compile-time, not in production - Policy templates: Auto-generate validation rules from CRD schemas - Compliance automation: Enforce organizational security and resource standards - Consistent patterns: Eliminate configuration drift

Outcome: Improved reliability and faster deployment cycles.

💻 For Application Developers¶

Challenge: Provisioning infrastructure requires deep Kubernetes expertise.

AMDF Value: - Simple interfaces: Use blueprints with only essential fields exposed - Type safety: IDE autocomplete and validation as you type - Clear examples: Auto-generated main.k files show exactly how to use resources - Self-service: Provision infrastructure independently

Outcome: Focus on application development, not infrastructure complexity.

Core Capabilities¶

🔍 Universal Discovery¶

• List all CRDs in your cluster with filtering
• Discover native Kubernetes resources (Pod, Service, Deployment, etc.)
• Support for any operator: Crossplane, Istio, ACK, KRO, custom CRDs

🏗️ Smart Generation¶

AMDF creates a complete toolkit for each resource:

1. Complete Schema (library/models/)
   • Full-fidelity, type-safe representation
   • All fields with documentation
   • For advanced use cases
2. Simple Blueprint (library/blueprints/)
   • Curated interface with essential fields
   • Sensible defaults
   • For day-to-day usage
3. Policy Template (library/policies/)
   • Auto-generated validation rules
   • Customizable compliance checks
   • Leverages KCL's native validation (we make it easier)
4. Usage Example (library/main.k)
   • Working code you can run immediately
   • Shows best practices
   • Includes policy application

🛡️ Validation Made Easy¶

While KCL provides powerful validation capabilities, AMDF makes them accessible:

• Auto-scaffolded policies: We generate policy templates from CRD schemas
• Common checks included: Security, resources, naming conventions
• Easy customization: Uncomment what you need, add your own rules
• Compile-time feedback: Fast validation during development

# AMDF generates this for you
schema DeploymentPolicyMixin:
    check:
        _replicas >= 2, "Minimum 2 replicas for HA"
        all container in _template.spec.containers {
            "latest" not in container.image
        }, "No 'latest' tags allowed"

🤖 AI-Native Experience¶

• MCP Server: Standardized interface for AI development tools
• Guided Mode: Interactive wizard with contextual assistance
• Local AI Integration: Offline assistance with Ollama

📦 Distribution & Reusability¶

• Package modules in OCI registries
• Share across teams and organizations
• Version control and dependency management
• GitOps-ready workflows

Architecture¶

AMDF follows a clean, modular design:

Complete Architecture¶

The diagram above shows the complete end-to-end flow:

• Platform Team uses AMDF CLI or MCP Server to generate KCL schemas from CRDs

• Core Logic Library produces type-safe schemas, blueprints, and policies

• Foundational Packages enable distribution via OCI Registry, Git, or YAML

• Product Teams/Devs consume packages locally via imports or through GitOps workflows (Argo + KCL Plugin)

• Multi-cloud Providers (AWS, GCP, Azure) receive validated configurations via Kubernetes

Core Engine Detail¶

graph TB
    subgraph "User Interfaces"
        CLI[CLI Commands]
        MCP[MCP Server]
        GUI[Guided Mode]
    end

    subgraph "Core Engine"
        DISC[CRD Discovery]
        GEN[Schema Generator]
        BP[Blueprint Creator]
        POL[Policy Scaffolder]
    end

    subgraph "Output"
        SCHEMA[Type-Safe Schemas]
        BLUE[Simple Blueprints]
        POLICY[Policy Templates]
        EXAMPLE[Usage Examples]
    end

    CLI --> DISC
    MCP --> DISC
    GUI --> DISC
    DISC --> GEN
    GEN --> BP
    GEN --> POL
    GEN --> SCHEMA
    BP --> BLUE
    POL --> POLICY
    GEN --> EXAMPLE

Getting Started¶

Installation¶

pip install amdf

Basic Workflow¶

# Discover what's available
amdf list-crds --filter aws
amdf list-k8s --filter deployment

# Generate schemas from CRDs
amdf generate instances.ec2.aws.upbound.io

# Generate schemas from native Kubernetes objects
amdf generate-k8s Pod
amdf generate-k8s Service

# Use the blueprints
import library.blueprints.Instance
import library.blueprints.Service

# Crossplane resource
server = Instance.InstanceBlueprint {
    _metadataName = "web-server"
    _instanceType = "t3.medium"
}

# Native Kubernetes resource
service = Service.ServiceBlueprint {
    _metadataName = "nginx"
    _namespace = "demo"
    _labels = {app = "nginx"}
    _ports = [{name = "http", port = 80, protocol = "TCP", targetPort = 80}]
    _selector = {app = "nginx"}
    _type = "ClusterIP"
}

Interactive Mode¶

# Guided experience with AI explanations
amdf guided --ai-model qwen3-coder:30b

Ecosystem Integration¶

AMDF orchestrates a best-in-class stack to deliver infrastructure as code:

• KCL (Configuration Engine) Provides the modeling language, validation logic, and schema generation capabilities.

• CNCF Ecosystem (Target Resource Model) Compatible with any Kubernetes CRD, like Crossplane, ACK, KRO, Config Connector, and custom operators.

• Kubernetes (Control Plane) Serves as the universal API and reconciliation engine that AMDF extends.

• Ollama (Local Intelligence) Powers the integrated AI assistant for offline schema explanations and guidance.

• Model Context Protocol (MCP) (Agentic Interface) Exposes AMDF as a standardized server, allowing AI clients to directly interact with, query, and generate infrastructure.