Griffin AI Agent Structure

The Griffin AI Agent framework is designed to facilitate the creation and execution of AI Agents through a combination of Agent Tasks and Tools. These components are integral to defining an Agent (Program), which represents the actual implementation of an agent instance executed within the Griffin AI physical architecture.

Figure 4: Structure of Griffin AI Agent Framework

Tools Framework

At the core of the AI Agent framework is the Tools Framework, a robust set of instruments grouped by domains and accessible to every AI Agent in real-time. These tools enable AI Agents to interact with the external world, including blockchain ecosystems, enhancing their capability to perform a wide range of tasks from simple computations to complex blockchain interactions.

AI Crew

For the execution of more complex tasks, AI Agents can be organized into a multi-agent group known as an AI Crew. Within this configuration, agents collaborate to accomplish a defined set of tasks through directed processes. Each AI Crew is designed to operate using "Processes," a platform entity that manages task execution by coordinating agent activities. This setup allows individual agents within the crew to function cohesively, streamlining their efforts to achieve common objectives efficiently.

Process Execution

The execution of tasks within an AI Crew is managed through processes, which can be visualized as a graph containing four essential components:

• Entry Point: The initiation phase of the process.

• Agents as Nodes: Each agent involved is represented as a node within the process graph.

• Orchestration Logic: Directed edges, including conditional edges, represent the orchestration logic guiding the interactions and sequences of tasks among the agents.

• End Point: The conclusion or output phase of the process.

Process Execution

Processes within the Griffin AI framework can be implemented in several forms to suit different operational needs, as detailed in the table below.

Name	Definition
Sequential	Tasks are executed one after the other, ensuring a systematic completion of tasks.
Hierarchical	Tasks are organized in a managerial hierarchy, where delegation and execution follow a structured chain of command. To enable this hierarchical process, a manager Agent must be specified within the crew, allowing the manager to create and manage tasks.
Consensual Process (Planned)	This process type aims for collaborative decision-making among agents regarding task execution, introducing a democratic approach to task management within Griffin.AI platform.

Table 9: Types of Process Implementation