Salami

Infrastructure As Natural Language

🤔 What is this?

Salami is a declarative domain-specific language for cloud infrastructure based on natural language descriptions. Salami compiler uses GPT4 to convert the natural language to Terraform code. You can think of Salami as writing documentation for each cloud resource object, and letting the compiler take care of converting that to IaC (Infrastructure as Code).

Short demo video | Release blog post

🚀 Getting Started

Installation

Homebrew (Mac OS, Linux):

brew tap petrgazarov/salami
brew install salami

Manual:

Download the latest binaries for Mac OS, Linux and Windows from the releases page.

Config

The root of your project should contain a salami.yaml config file.

Example:

compiler:
  target:
    platform: terraform
  llm:
    provider: openai
    model: gpt4
    api_key: ${OPENAI_API_KEY}
  source_dir: salami
  target_dir: terraform

Configuration Setting	Description	Required
compiler.target.platform	Platform to target. Only terraform value is currently supported.	Yes
compiler.llm.provider	Provider for the LLM. Only openai value is currently supported.	Yes
compiler.llm.model	Model used by the provider. Only gpt4 value is currently supported.	Yes
compiler.llm.api_key	OpenAI API key. To set it to an env variable, use the ${ENV_VAR} delimeter.	Yes
compiler.llm.max_concurrent	Maximum number of concurrent API calls to OpenAI API. Default is 5.	No
compiler.source_dir	The directory where your Salami files are located.	Yes
compiler.target_dir	The directory where the Terraform files should be written.	Yes

Usage

From the root of your project, run:

salami compile

For verbose output, run:

salami -v compile

🎨 Design

Constructs

Salami files are mostly comprised of natural language, with several special constructs:

1. Blocks - multiline blocks of text that each represent either a resource or a variable; delimited by double newlines.
2. Constructor functions - functions that are used to specify the nature of the block; start with an @ symbol.
3. Variable references - references to variables that are defined in the program; delimited by curly braces.
4. Resource references - references to resources that are defined in the program; start with a dollar sign.

Example Salami code with 3 blocks: VPC resource, Security Group resource and the container_port variable:

For more examples, see the examples directory. Each example has a README file with instructions on how to run it.

Constructor function signatures

@resource

Position	Argument	Type	Required?	Examples
1	resource type	string	Yes	aws.s3.bucket, AWS S3 Bucket
2	logical name	string	Yes	ApiCluster, prod_bucket_1

@variable

Position	Argument	Type	Required	Examples
1	name	string	Yes	container_port, logs_bucket_name
2	variable type	string	Yes	string, number, boolean
3	default	any	No	8080, logs_bucket_1fdretbnHUdfn

Lock file

The compiler generates a lock file that includes parsed Salami objects and the resulting Terraform code. It is used to determine which objects have changed since the last compilation. Unchanged objects are not sent to LLM, which makes the compilation process much faster.

File extension

.sami is the extension for Salami files.

✅ VS Code Extension

It's recommended to install the Salami VS Code extension. It provides highlighting functionality for the .sami files.

😍 Contributing

Contributions are welcome! For non-trivial contributions, please open an issue first to discuss the proposed changes.