AI Infrastructure Agent

The AI Infrastructure Agent is an intelligent system designed to manage AWS infrastructure using natural language commands. Instead of writing complex infrastructure-as-code, users can describe their desired state in plain English. The agent maintains a Terraform-like state to ensure accuracy and includes conflict detection and dry-run modes to provide safety during infrastructure changes.

Key Capabilities:

• Natural Language Interface: Describe infrastructure requirements directly.
• Multi-AI Provider Support: Compatible with OpenAI, Google Gemini, Anthropic, AWS Bedrock Nova, and Ollama (for local LLM execution).
• AWS Resource Management: Supports VPC, EC2, Security Groups, Auto Scaling Groups, and ALBs.
• Web Dashboard: Provides a visual interface for managing infrastructure.

The project uses a sophisticated mock system to provide realistic testing environments without hitting live AWS endpoints. The testing architecture integrates real production logic (like pattern matching and field resolution) with the following mock components:

• MockMCPServer: Simulates a complete MCP server with 50+ tools.
• MockStateManager: Simulates state management including real resource analysis.
• MockAWSClient: Simulates AWS API responses.
• MockAdapters: Provides service-specific mocks for isolated testing.

The agent operates through a multi-step lifecycle to ensure safe infrastructure management:

1. AI Analysis & Planning: The agent receives a natural language request, retrieves the current infrastructure state, queries available tools via the MCP Server, and generates an execution plan using an LLM.
2. Plan Validation: The plan is validated against AWS (dry-run checks) to ensure feasibility.
3. User Approval: The agent presents the detailed plan to the user via the Web Dashboard. No changes are made until the user approves.
4. Execution & Monitoring: Upon approval, the agent executes the plan, creating resources in the correct dependency order and monitoring progress in real-time.

If the AI agent encounters an error during infrastructure deployment (e.g., due to environment mismatches or AWS state conflicts), it features a built-in recovery mechanism.

When an error occurs, the agent automatically triggers a recovery plan to analyze the failure and attempt to fix the issue. Users can review the recovery steps and choose to either let the agent proceed with the fix or skip the problematic step if the core infrastructure is already in the desired state.

The Discovery System is responsible for scanning and mapping existing infrastructure. Its core capabilities include:

• Multi-Service Scanning: Comprehensive discovery across various AWS services.
• Relationship Mapping: Identifying and mapping the relationships between discovered resources.

Note: State Correlation and Drift Detection are currently marked as 'Not Ready'.

When using the AI Infrastructure Agent to manage cloud resources, follow these core principles to ensure safety and accuracy:

1. Use Natural Language: Describe your infrastructure needs in plain English.
2. Be Specific for Consistency: Different AI models interpret requests differently; detailed prompts reduce ambiguity.
3. Review Before Executing: Always carefully inspect the generated infrastructure plan before allowing the agent to apply changes.
4. Verify in AWS Console: After execution, confirm that the provisioned resources match your expectations in the AWS Management Console.
5. Follow Security Best Practices: Especially when defining security groups and network access for production environments.

The system uses a two-tier approach to manage AWS resources:

1. Adapters Layer: Implements the Adapter pattern to provide a consistent interface across different AWS services. Every service-specific adapter (e.g., VPC Adapter, EC2 Adapter) implements the AWSResourceAdapter interface, ensuring standardized resource lifecycle management, validation, and metadata extraction.
2. AWS Integration Layer: Provides the low-level abstraction for the AWS SDK. It manages multi-service clients, handles authentication (IAM roles/credentials), and implements automatic retries with exponential backoff.

type AWSResourceAdapter interface {
    GetResourceType() string
    ValidateResource(resource *types.ResourceState) error
    NormalizeResource(rawResource interface{}) (*types.ResourceState, error)
    ExtractMetadata(resource interface{}) map[string]interface{}
}

The AI Infrastructure Agent follows a four-step lifecycle to transform natural language into AWS infrastructure:

1. Enter Request: Describe your infrastructure needs in plain English via the Web UI.
2. Review Plan: The AI generates a detailed execution plan. Review this for dependencies, parameters (like instance types or AMIs), and resource IDs.
3. Approve & Execute: Once you click "Confirm & Execute", the agent performs the AWS API calls sequentially.
4. Verify Resources: Check the dashboard or the AWS Console to confirm the resources were created correctly.

The Agent Layer is the intelligence core of the system. It uses a StateAware Agent to bridge LLMs with infrastructure via the Model Context Protocol (MCP).

Key capabilities include:

• Multi-AI Provider Support: Integrates with OpenAI, Google Gemini, Anthropic Claude, AWS Bedrock Nova, and Ollama.
• Request Processing: Converts natural language into context-aware execution plans with dependency resolution.
• Plan Execution: Manages step-by-step execution, ensuring resources are created in the correct dependency order. It supports Dry-Run mode to simulate changes without actual resource creation.
• Resource Analysis: Uses a Pattern Matcher, Field Resolver, and Value Type Inferrer to map AWS API responses to internal structures.
• Safety: Performs pre-execution validation to detect conflicts and ensure consistency.

The Resource Analysis System enables intelligent handling of AWS resources through three primary mechanisms:

1. Pattern Matching Engine: Uses compiled regex patterns to detect AWS resource types and automatically maps tool names to specific resource types. It is configuration-driven and optimized via pattern caching.
2. Field Resolution Engine: Performs runtime mapping between AWS API responses and internal data structures. It supports deep object traversal using dot notation (e.g., parent.child.field) and includes automatic type conversion and error recovery for missing fields.
3. Value Type Inference: Uses context-aware rules and pattern matching based on field names and descriptions to intelligently infer data types.