Supervisor Agent

The Supervisor Agent is a specialized AI agent in Lamatic.ai designed to manage and coordinate multi-agent flow. It serves as the central hub, collecting input, maintaining structured memory, and orchestrating the execution of sub-agents based on predefined logic. This ensures an efficient, iterative process for dynamic AI-powered applications such as structured decision-making systems and task automation.

Why Use the Supervisor Agent?

• Orchestrate Complex AI Flow: Manage multiple agents dynamically, ensuring structured execution.
• Maintain Context and Memory: Retain past interactions to improve continuity in multi-step tasks.
• Enable Adaptive Execution: Define agent paths, loop conditions, and AI-generated decision trees.

Key Features

How to Use the Supervisor Agent?

Creating a Supervisor Agent via Flow Editor

1. Add an Agent Node - Select Supervisor Agent from the node list.
2. Define Agent Paths - Set up branches for different sub-agents.
3. Configure Prompt Templates - Customize how the agent interacts with users and sub-agents.
4. Select an LLM Model - Choose the AI model for text generation.
5. Set Loop & Stop Conditions - Define execution limits to prevent infinite loops.
6. Deploy & Connect - Integrate the agent into the flow and execute the workflow.

Creating a Supervisor Agent via Agent Dashboard

1. Navigate to the Agents Page - Click "New Agent."
2. Select Supervisor Agent - Choose it from the available agent types.
3. Configure Agent Settings - Adjust memory, iteration limits, and agent paths.
4. Deploy the Agent - Save and activate it in your workflow.

Configuration Options

Agent Path Configuration

The Supervisor Agent allows you to define different execution paths for specialized sub-agents. Each agent path represents a distinct functionality or responsibility within your workflow.

Configuring Agent Paths

For each agent path, you'll need to define three key components:

1. Agent Name: A descriptive identifier for the sub-agent's role (e.g., "Researcher", "Writer", "Editor")
2. Agent Description: A clear explanation of the sub-agent's purpose and responsibilities
3. Agent Schema: A structured JSON schema defining the expected output format from the agent

Defining Agent Schema

The agent schema uses a JSON structure to specify the format and requirements for the agent's output. In the example shown:

{
  "research": {
    "type": "string",
    "required": true,
    "description": "This implies the final research fetched by the research team, which is supposed to be passed to the writing team"
  }
}

This schema defines:

• A field named research that will store the agent's output
• The data type is specified as string
• The field is marked as required (true)
• A description that explains the purpose and usage of this data

Visual Schema Builder

Lamatic.ai provides a visual schema builder to easily define your agent path configurations:

1. Field Name: Enter the output field name (e.g., "research")
2. Data Type: Select the appropriate type (string, number, boolean, etc.)
3. Required: Toggle whether the field is mandatory
4. Description: Add explanatory text to clarify the field's purpose

By properly configuring your agent paths, you enable the Supervisor Agent to effectively route tasks, maintain context, and orchestrate complex flow across multiple specialized agents.

Save Agent Configuration

You can save the configuration of any agent by clicking on the Load Save Config button and selecting Save as New. This will save the configuration of the agent and you can use it later by clicking on the Load Configuration button in other agents.

Low-Code Example

nodes:
  - nodeId: agentNode_960
    nodeType: agentNode
    nodeName: Supervisor
    values:
      tools: []
      agents:
        - name: Research
          schema: {}
          description: Agent 1 description
        - name: Writer
          schema: {}
          description: Agent 2 description
      prompts:
        - id: 9ed337ec-a4c0-4885-9066-f2853347d20b
          role: system
          content: >-
            Research on the topic {{triggerNode_1.output.topic}} and pass on
            your search to the writer to generate the ideal text.
      messages: '[]'
      stopWord: 'end'
      connectedTo: agentLoopEndNode_242
      maxIterations: 5
      generativeModelName:
        type: generator/text
        model_name: gpt-4o-mini-2024-07-18
        credentialId: f6930a0d-9947-472c-b64b-2244a131cad0
        provider_name: openai
        credential_name: OPEN_AI_KEY
    needs:
      - triggerNode_1
      - agentLoopEndNode_242
    connections:
      - condition: Research
        value: webSearchNode_852
        type: conditionEdge
        invisible: false
      - condition: Writer
        value: LLMNode_238
        type: conditionEdge
        invisible: false
      - condition: Agent Loop End
        value: agentLoopEndNode_242
        type: agentLoopEdge
        invisible: true

Output

output

• A nested object containing the output generated by the Supervisor Node. This depends and modfies based on the paths and agents defined in the configuration.

Example Output

{
  “output”: {
    "Research": {
      "output": "Research on the topic Bali and pass on your search to the writer to generate the ideal text."
    },
    "Writer": {
      "output": "The text generated by the writer based on the research is..."
    }
  }
}

Troubleshooting

Common Issues

Debugging

1. Check Lamatic Flow logs for error details.
2. Verify API Key.