Core concepts

Sublayer Generators

Generators are responsible for generating specific outputs based on input data. They focus on a single generation task and do not perform any actions or complex decision-making. Generators are the building blocks of the Sublayer framework.

llm_output_adapter

Output adapters in Sublayer are powerful components that define the structure and format of the AI-generated output. They serve as a bridge between the raw AI responses and the structured data your application expects.

The framework comes with a set of built-in output adapters for common use-cases, as well as the ability to create custom output adapters for more specialized ones.

To learn more about creating your own custom output adapters, check out the Custom Components: Output Adapters page.

The name, description, and structure defined in the output adapter is passed to the LLM to provide a structure for the output that the generator will produce. This is useful when building applications that make calls to LLMs so that you can be sure of the structure of the output that you will receive.

The current built in output adapters are:

Single String

:single_string is used when you just want a single, specific string back from the LLM. For example, if you want the LLM to generate a specific piece of code for you, or a description of a piece of code.

Usage:

llm_output_adapter type: :single_string,
  name: "code_description",
  description: "A description of what the code does, its purpose,functionality, and any noteworthy details"

Examples:

List of Strings

:list_of_strings is used when you want the LLM to return you a list of strings for a specific topic. For example if you want the LLM to generate a list of keywords from a text, or a list of suggestions for a specific topic.

llm_output_adapter type: :list_of_strings,
  name: "suggestions",
  description: "List of keyword suggestions"

Examples:

String Selection From List

:string_selection_from_list is used when you want the LLM to select a specific option from a list that you give it. For example if you give the LLM a comment from social media and you want to give it a sentiment grade or if you have a list of options and you want the LLM to select one of them given some user action.

  llm_output_adapter type: :string_selection_from_list,
    name: "sentiment_value",
    description: "A sentiment value from the list",
    options: -> { @sentiment_options }

**Examples:##

Named Strings

:named_strings is used when you want to receive an object back with a specific set of string attributes. For example if you give the LLM a blog post, you can have it return a twitter blurb, a linkedin blurb, and a short summary of the post.

llm_output_adapter type: :named_strings,
  name: "blog_post_blurbs",
  description: "A set of social media blurbs for the given blog post",
  attributes: [
    { name: "twitter_blurb", description: "A 2 sentence blurb for sharing the blog post on twitter" },
    { name: "linkedin_blurb", description: "A blurb for the blog post for sharing on LinkedIn" },
    { name: "short_description", description: "A short summary of the blog post" }
  ]

Examples:

Prompt

This method is required for all generators and is used to define the input that we'll send to the model. In the examples below we use heredoc syntax for the whole method, but as long as the method returns a string, you can use any mechanism you'd like.

Example:

def prompt
  <<~PROMPT
    Please provide a description of the code you would like to generate.
  PROMPT
end

Examples

Generate your own generator:

Previous
Triggers
Next
Actions