🚀 SkyBot

Create resource on the cloud with natural language using AI-powered Terraform generation

📖 Features

• Natural language-based resource creation
• Support for AWS cloud resources (S3 buckets, EC2 instances, etc.)
• Local infrastructure development using LocalStack
• Component-based infrastructure management
• Interactive chat interface for cloud resources
• Support for multiple infrastructure components
• Self-healing infrastructure creation with automatic error fixing

🛠️ Prerequisites

• Python 3.10 or higher
• Required packages (to be installed via pip):

  pip install skybot
  

• Terraform installed:

  brew install terraform
  

• AWS CLI configured:

  aws configure
  

P.S, make sure to configure the default region as well.

• OpenAI API key:

  export OPENAI_API_KEY='your_api_key_here'
  

• For local development:

  docker pull localstack/localstack
  docker run -d -p 4566:4566 localstack/localstack
  

📚 Command Structure

Initialize a new project:

skybot init [--verbose] [--local]

Create a new component:

skybot component create --prompt "Your infrastructure description" --name component-name [--verbose] [--force] [--model MODEL_NAME] [--self-healing] [--max-attempts N] [--keep-on-failure]

Delete all components:

skybot component delete [--force]

Destroy all components infrastructure:

skybot component destroy [--force]

Edit a component:

skybot component edit component-name

Chat about your infrastructure:

skybot chat component-name

Check SkyBot version:

skybot version

📊 Usage Examples

1. Initialize a new project:

skybot init

2. Create a web server component with self-healing:

skybot component create --prompt "Create an EC2 instance with nginx installed" --name web-server --self-healing

3. Create a local S3 bucket for testing:

skybot component create --prompt "Create an S3 bucket" --name test-bucket --local

4. Create a database component with custom retry attempts:

skybot component create --prompt "Set up an RDS instance for PostgreSQL" --name database --self-healing --max-attempts 5

5. Chat about your infrastructure:

skybot chat web-server

🗂️ Project Structure

When you initialize a project, SkyBot creates a .skybot directory with the following structure:

.skybot/
└── default/
    ├── backend.tf
    ├── provider.tf
    ├── component1.tf
    ├── component2.tf
    └── ...

Each component is stored as a separate Terraform file in the workspace directory.

🔧 Advanced Features

Self-Healing Infrastructure Creation

SkyBot includes a self-healing feature that automatically fixes Terraform errors during resource creation:

• Enable with --self-healing flag
• Set maximum retry attempts with --max-attempts N (default: 3)
• Uses AI to analyze errors and fix configuration issues
• Maintains original infrastructure intent while resolving dependencies
• Shows detailed fix explanations for transparency
• Use --keep-on-failure to preserve generated Terraform files even when errors occur (useful for debugging)

Example with self-healing:

skybot component create \
  --prompt "Create a highly available EC2 setup with auto-scaling" \
  --name ha-web \
  --self-healing \
  --max-attempts 5 \
  --keep-on-failure

If Terraform encounters errors during plan or apply:

1. SkyBot analyzes the error output
2. AI suggests fixes while preserving the original intent
3. Retries the operation with fixed configuration
4. Continues until success or max attempts reached
5. If --keep-on-failure is set, preserves the generated Terraform files for inspection even if errors occur

Langfuse Monitoring

SkyBot supports observability and monitoring of AI interactions through Langfuse:

• Set up Langfuse credentials:

  export LANGFUSE_PUBLIC_KEY='your_public_key'
  export LANGFUSE_SECRET_KEY='your_secret_key'
  

• All AI interactions are automatically logged to your Langfuse dashboard

Alternative Models

SkyBot supports multiple AI models for infrastructure generation through LiteLLM integration. While OpenAI is the default provider, you can use other models by setting the appropriate API key and specifying the model:

Note: Even when using alternative models, the OPENAI_API_KEY environment variable is still required for certain auxiliary tasks within SkyBot.

Using Groq Models

export GROQ_API_KEY='your_api_key'
skybot component create \
  --name eks-cluster-1 \
  --prompt "create an EKS cluster named MyKubernetesCluster" \
  --self-healing \
  --model "groq/deepseek-r1-distill-llama-70b"

Using Perplexity Models

export PERPLEXITY_API_KEY='your_api_key'
skybot component create \
  --name eks-cluster-1 \
  --prompt "create an EKS cluster named MyKubernetesCluster" \
  --self-healing \
  --model "perplexity/sonar"

The --model flag allows you to specify which model to use for infrastructure generation. Make sure to set the corresponding API key as an environment variable before running the command.

SkyBot supports all models available through LiteLLM (see LiteLLM Documentation), including but not limited to:

• OpenAI (default), for instance: gpt-4o, o3-mini
• Groq, for instance: groq/deepseek-r1-distill-llama-70b
• Perplexity, for instance: perplexity/sonar-pro
• Anthropic, for instance: anthropic/claude-3-5-sonnet
• Google VertexAI
• AWS Bedrock
• Azure OpenAI
• Hugging Face
• And many more

Each provider requires its own API key to be set as an environment variable. Common examples:

• OPENAI_API_KEY for OpenAI models (required for all setups)
• GROQ_API_KEY for Groq models
• PERPLEXITY_API_KEY for Perplexity models
• ANTHROPIC_API_KEY for Anthropic models
• AZURE_API_KEY for Azure OpenAI models

Refer to the LiteLLM documentation for the complete list of supported models and their corresponding environment variables.