Orchestration

CloudThinker Multi-Agent Orchestration: How It Works

CloudThinker’s orchestration engine leverages a network of specialized agents to deliver intelligent, adaptive workflow automation. This architecture is designed for flexibility, scalability, and precise task delegation across complex business processes.

Network Multi-Agent Systems

Network Multi-Agent Systems

At its core, CloudThinker utilizes a multi-agent system-a collection of autonomous, role-specific agents that collaborate to achieve common goals. Each agent can be an LLM, an API integration, a serverless function, or a custom logic component, and is assigned a unique role and description. The orchestrator intelligently routes user requests to the most appropriate agent, ensuring context-aware processing and seamless handoff between agents

Workflow Control with Mentions and Hashtags

CloudThinker introduces a powerful, intuitive way to control and direct workflows using mentions (@) and hashtags (#):

• Mentions (@agent): Directly address or assign tasks to a specific agent. For example, @scheduler can trigger the scheduling agent to handle a booking request.

• Hashtags (#topic): Tag or categorize requests, enabling agents to recognize context or trigger predefined workflows. For example, #push_alert can push the alert in the workflow. User can use hashtags to proactive control the way agents use tools, documents, resources

This approach allows users and agents to dynamically influence the orchestration flow, making the system highly adaptable and user-friendly.

Role-Specific Agents and Orchestration Logic

Each agent in CloudThinker is designed for a specific role-such as scheduling, data enrichment, or customer support. The orchestration process follows these key steps:

1. Request Initiation: The user submits a request, potentially using @ mentions or # hashtags to guide the workflow.

2. Classification: An intelligent classifier analyzes the request, agent descriptions, and ongoing conversation context to determine which agent (or agents) should handle the task

   1. Agent Selection: The orchestrator selects the appropriate agent(s) based on role, context, and workflow directives.

   2. Request Routing: The input is routed to the selected agent, which processes the task and may delegate subtasks to other agents as needed.

   3. Context Management: The orchestrator maintains a comprehensive conversation and task history, ensuring continuity and coherence across all interactions

3. Response Delivery (Group messages): The orchestrator compiles and delivers the final response to the user, updating all relevant histories.

Advanced Coordination: SupervisorAgent (@anna)

For complex workflows involving multiple agents, CloudThinker employs a SupervisorAgent-a lead agent responsible for:

• Coordinating parallel or sequential execution of subtasks among specialized agents

• Maintaining global context and ensuring coherent, unified responses

• Dynamically delegating tasks based on real-time workflow requirements

This enables sophisticated team-based problem-solving, such as customer support with specialized sub-teams, project management, or multi-step data processing.

Key Benefits

• Scalability: Easily add or customize agents to fit evolving business needs

• Flexibility: Handle a wide range of workflows, from simple queries to complex, multi-step processes.

• Transparency: Use @ and # controls for clear, auditable workflow management.

• Seamless Context: Maintain conversation and task continuity across all agents and sessions

Summary Table: Key Orchestration Features

Feature	Description
Network Multi-Agent	Multiple specialized agents collaborate in a unified system
Mentions (@)	Direct task assignment to specific agents
Hashtags (#)	Tagging/categorization for context-aware workflow control
Role-Specific Agents	Each agent has a defined role, description and instruction
SupervisorAgent	Advanced coordination, parallel tasking, and dynamic delegation
Context Management	Persistent, cross-agent conversation (Group messages) and workflow history (Conversations)

CloudThinker’s orchestration framework empowers organizations to automate and optimize complex workflows with clarity, efficiency, and adaptability-driven by the seamless collaboration of intelligent, role-specific agents