AgentK is an autonomous agent AGI system with self-evolving and modular characteristics. The "K" in its name stands for kernel, meaning small core. The design goal is to make AgentK the minimal set of agents and tools needed to bootstrap itself, and then gradually grow its own mind. AgentK's mind is composed of two parts: agents and tools. The agents collaborate with one another to solve problems, while the tools enable the agents to interact with the external world. Both are developed as ordinary Python files and stored in the agents and tools directories, so it is very easy to track progress and also convenient for users to contribute code themselves.
The agents that make up the kernel include: Hermes, as the orchestrator, responsible for interacting with humans to understand goals, managing the creation and assignment of tasks, and coordinating the activities of other agents; AgentSmith, as the architect, responsible for creating and maintaining other agents, ensuring that agents are equipped with the necessary tools and testing their functionality; ToolMaker, as the tool developer within the system, creating and optimizing the tools needed by agents to perform tasks, ensuring that the system remains flexible and well equipped; WebResearcher, as the knowledge collector, performing in-depth online research to provide the system with the latest information, enabling agents to make informed decisions and execute tasks effectively.
AgentK is encouraged to write tests itself. Further improvement is currently underway in detecting and fixing agents and tools with abnormal behavior, and this work is still in progress. AgentK is built on the LangGraph and LangChain frameworks. It runs in an isolated Docker container, so the latest version of Docker needs to be installed on the system. The usage method is: copy .env.template to .env, and then set the environment variables in .env.
