Multi-Agent Structure

Multi Agent Structure

Multi-agent systems in Knovvu Virtual Agent enable sophisticated workflows where specialized agents collaborate to handle complex scenarios. This structure enables Knovvu Virtual Agent to handle complex, enterprise-scale deployments while maintaining flexibility, reliability, and performance. The modular design allows organizations to start with simple single-agent implementations and scale to sophisticated multi-agent orchestrations as their needs evolve.

Supervisor

The Supervisor agent acts as the orchestrator in multi-agent systems, managing the overall workflow and coordinating between specialized agents. It serves as the primary decision-maker for task delegation and result synthesis.

• Receives initial requests and determines handling strategy. Identifies required specialized agents and delegates subtasks based on capabilities. In cases where delegation is not required, it can also execute the task itself.

Single Agent

Single agents are specialized units designed to handle specific domains or task types. Each agent operates autonomously within its defined scope.

• Different agents can specialize in various domains such as billing and payment processing, technical support and troubleshooting, product information and recommendations, appointment scheduling and management, or data analysis and reporting.

• Each single agent maintains deep knowledge in its specialized domain. It has access to specific tools and integrations relevant to its function. The agent can handle complete workflows within its scope and maintains its own context and memory for domain-specific information.

Sub-Agent

Sub-agents are focused components that handle specific subtasks within a larger workflow. They're typically invoked by supervisors to perform discrete operations. Once invoked, the sub-agent interacts directly with the end user until the task is completed. If the user’s request falls outside the sub-agent’s capabilities, it escalates the conversation back to the supervisor.

• Sub-agents are designed for the execution of specific tasks. These agents are highly reusable across different workflows and optimized for specific operations like data validation, format conversion, or API interactions. For example, in an appointment scheduling scenario, one sub-agent might collect the address while another manages the scheduling.

• All agents within the supervisor architecture share full state and conversation history, ensuring conversation continuity. Information provided by the user can be utilized by any sub-agent when needed. They work collaboratively toward a single goal while handling different tasks.

• Sub-agents are composed of single agents. A single agent can also operate under a supervisor as a sub-agent, allowing collaboration and coordination through defined collaboration instructions.