CAS is a multi-agent framework that allows you to create and interact with personalities or agents, each with their own unique behaviors and responses. The framework currently supports three predefined personalities: Aegis, Cognita, and Sophia. You can also dynamically generate new agents based on your needs. CAS was written using prompt engineering techniques to be interpreted by the LLM itself through the system prompt. To the LLMs, CAS appears to be a program therefore it is treated as such. The LLMs will read the comments and interpret them as commands as well reinforcing the behavior of the LLM.
CAS is designed for use as an LLM System Prompt, the framework allows you to:
- Define static personalities with distinct behaviors and responses.
- Dynamically generate new temporary agents on demand.
- Enable conversational communication between agents.
You can find documentation on how to use and customize CAS in CAS.md.
Additional resources are available under the docs directory.
Under the src directory, you will find files for each version of CAS.
Under the templates directory, you will find CAS included with various prompt templates applied.
- Read the documentation so you know what to expect.
- Choose the corresponding system prompt for the LLM from the template directory. (Using the wrong prompt template will result in strange behavior or errors.)
- Run the system with your desired prompt style.
- Chat.
We welcome contributions! Please follow these guidelines:
- Write an SOP for tests to validate your changes using the provided test template found in the
testsdirectory - Follow best practices and adhere to our contributing guide
- Create new agents or personalities in the appropriate directories under `src/prompts/[your_prompt]
CAS is licensed under the (TBD). See LICENSE for more information.