List view
This milestone focuses on replacing the existing Autogen dependency with a custom, robust, and scalable solution for interacting with Large Language Models. The core deliverable is a new client module that serves as a single, centralized point of access for all agents. The client will be designed to handle multiple local and external models, manage GPU resource contention, and ensure a consistent and reliable experience for all agents. The use of Pydantic, Abstract Base Classes (ABCs), and the Singleton pattern will enforce a strict, maintainable, and extensible architecture. By the end of this milestone, all agents will be refactored to communicate with this new client, and the framework will be ready to scale by adding new models and API providers with minimal effort. This is a critical step towards improving performance, reducing external dependencies, and establishing a stable foundation for the future of our multi-agent system.
No due date•6/6 issues closedThis proposal outlines the architecture for a new, custom GroupChat module to replace the dependency on AutoGen's GroupChat and GroupChatManager. This new module will be built to be flexible and reusable for all multi-agent conversation and debate functions within the TwinRAD framework.
No due date•7/7 issues closed