Supervisor

​

Node Details

• Name: Supervisor
• Type: Supervisor
• Category: Multi Agents
• Version: 3.0

​

Base Classes

• Supervisor
• (Additional base classes from Supervisor implementation)

​

Parameters

​

Inputs

1. Supervisor Name
   • Type: string
   • Default: “Supervisor”
   • Description: The name of the supervisor agent
2. Supervisor Prompt
   • Type: string
   • Description: Custom prompt for the supervisor, must contain {team_members}
   • Default: System-provided prompt template
3. Tool Calling Chat Model (Required)
   • Type: BaseChatModel
   • Description: Language model used for decision-making (must support function calling)
   • Compatible Models:
     • ChatOpenAI
     • ChatMistral
     • ChatAnthropic
     • ChatGoogleGenerativeAI
     • GroqChat
4. Agent Memory (Optional)
   • Type: BaseCheckpointSaver
   • Description: Used to save and restore the agent’s state
5. Summarization (Optional)
   • Type: boolean
   • Description: Enables final output as a conversation summary
6. Recursion Limit
   • Type: number
   • Default: 100
   • Description: Maximum number of recursive calls allowed
7. Input Moderation (Optional)
   • Type: Moderation[]
   • Description: Filters for detecting potentially harmful text

​

Functionality

1. Coordination
   • Manages worker agent selection
   • Orchestrates conversation flow
   • Monitors task progress
   • Optimizes task completion
2. Decision Making
   • Analyzes conversation state
   • Selects appropriate workers
   • Provides worker instructions
   • Ensures task coherence
3. Process Management
   • Handles conversation initialization
   • Manages state transitions
   • Controls recursion depth
   • Provides optional summarization

​

Use Cases

1. Project Management
   • Coordinate multiple specialists
   • Break down complex tasks
   • Monitor progress
   • Ensure goal alignment
2. Problem Solving
   • Distribute subtasks
   • Combine expert knowledge
   • Synthesize solutions
   • Maintain solution quality
3. Research Coordination
   • Assign research tasks
   • Manage information gathering
   • Validate findings
   • Compile results

​

Integration Notes

• Best performance achieved with GPT-4
• Requires function-calling capable models
• Supports multi-modal inputs for compatible models
• Special handling for various model types

​

Best Practices

1. Configuration
   • Set clear supervisor prompts
   • Choose appropriate models
   • Configure memory settings
   • Set reasonable recursion limits
2. Monitoring
   • Track conversation progress
   • Monitor worker performance
   • Watch for recursion issues
   • Evaluate task completion
3. Optimization
   • Fine-tune prompts
   • Adjust memory settings
   • Optimize model selection
   • Balance task distribution