Sublayer

A model-agnostic Ruby AI Agent framework. Provides base classes for building Generators, Actions, Tasks, and Agents that can be used to build AI powered applications in Ruby.

For more detailed documentation visit our documentation site: https://docs.sublayer.com.

Pre-1.0 we anticipate many breaking changes to the API. Our current plan is to keep breaking changes to minor, 0.x releases, and patch releases (0.x.y) will be used for new features and bug fixes.

To maintain stability in your application, we recommend pinning the version of Sublayer in your Gemfile to a specific minor version. For example, to pin to version 0.0.x, you would add the following line to your Gemfile:

gem 'sublayer', '~> 0.0'

Install the gem by running the following commands:

$ gem install sublayer

Or add this line to your application's Gemfile:

gem 'sublayer', '~> 0.0'

Sublayer is model-agnostic and can be used with any AI model. Below are the

Expects you to have an OpenAI API key set in the OPENAI_API_KEY environment variable.

Visit OpenAI to get an API key.

Usage:

Sublayer.configuration.ai_provider = Sublayer::Providers::OpenAI
Sublayer.configuration.ai_model = "gpt-4-turbo-preview"

Expects you to have a Gemini API key set in the GEMINI_API_KEY environment variable.

Visit Google AI Studio to get an API key.

Usage:

Sublayer.configuration.ai_provider = Sublayer::Providers::Gemini
Sublayer.configuration.ai_model = "gemini-pro"

Expect you to have a Claude API key set in the ANTHROPIC_API_KEY environment variable.

Visit Anthropic to get an API key.

Usage:

Sublayer.configuration.ai_provider = Sublayer::Providers::Claude
Sublayer.configuration.ai_model ="claude-3-opus-20240229"

Expects you to have a Groq API key set in the GROQ_API_KEY environment variable.

Visit Groq Console to get an API key.

Usage:

Sublayer.configuration.ai_provider = Sublayer::Providers::Groq
Sublayer.configuration.ai_model = "mixtral-8x7b-32768"

If you've never run a local model before see the Local Model Quickstart below. Know that local models take several GB of space.

The model you use must have the ChatML formatted v1/chat/completions endpoint to work with sublayer (many models do by default)

Usage:

Run your local model on http://localhost:8080 and then set:

Sublayer.configuration.ai_provider = Sublayer::Providers::Local
Sublayer.configuration.ai_model = "LLaMA_CPP"

Instructions to run a local model

1. Setting up Llamafile

cd where/you/keep/your/projects
git clone git@github.com:Mozilla-Ocho/llamafile.git
cd llamafile

Download: https://cosmo.zip/pub/cosmos/bin/make (windows users need this too: https://justine.lol/cosmo3/)

# within llamafile directory
chmod +x path/to/the/downloaded/make
path/to/the/downloaded/make -j8
sudo path/to/the/downloaded/make install PREFIX=/usr/local

You can now run llamfile

2. Downloading Model

click here to download Mistral_7b.Q5_K_M (5.13 GB)

3. Running Llamafile with a model

llamafile -ngl 9999 -m path/to/the/downloaded/Hermes-2-Pro-Mistral-7B.Q5_K_M.gguf --host 0.0.0.0 -c 4096

You are now running a local model on http://localhost:8080

llamafile -ngl 9999 -m Hermes-2-Pro-Mistral-7B.Q5_K_M.gguf --host 0.0.0.0 --nobrowser -c 2048 --gpu APPLE -t 12

run sysctl -n hw.logicalcpu to see what number to give the -t threads option

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.

Examples (in the /lib/sublayer/generators/examples directory):

• CodeFromDescriptionGenerator: Generates code based on a description and the technologies used.
• DescriptionFromCodeGenerator: Generates a description of the code passed in to it.
• CodeFromBlueprintGenerator: Generates code based on a blueprint, a blueprint description, and a description of the desired code.

Actions are responsible for performing specific operations to get inputs for a Generator or based on the generated output from a Generator. They encapsulate a single action and do not involve complex decision-making. Actions are the executable units that bring the generated inputs to life.

Examples:

• SaveToFileAction: Saves generated output to a file.
• RunCommandLineCommandAction: Runs a generated command line command.

Tasks combine Generators and Actions to accomplish a specific goal. They involve a sequence of generation and action steps that may include basic decision-making and flow control. Tasks are the high-level building blocks that define the desired outcome.

Examples:

• ModifyFileContentsTask: Generates new file contents based on the existing contents and a set of rules, and then saves the new contents to the file.

Agents are high-level entities that coordinate and orchestrate multiple Tasks to achieve a broader goal. They involve complex decision-making, monitoring, and adaptation based on the outcomes of the Tasks. Agents are the intelligent supervisors that manage the overall workflow.

Examples:

• CustomerSupportAgent: Handles customer support inquiries by using various Tasks such as understanding the customer's issue, generating appropriate responses, and performing actions like sending emails or creating support tickets.

There are sample Generators in the /examples/ directory that demonstrate how to build generators using the Sublayer framework. Alternatively below are links to open source projects that are using generators in different ways:

• Blueprints - An open source AI code assistant that allows you to capture patterns in your codebase to use as a base for generating new code.

• Clag - A ruby gem that generates command line commands from a simple description right in your terminal.

TBD

TBD