Skip to content

Multi Agent Inference

In the Multi Agent Inference setup, we offer a Human-in-the-Loop option. This feature allows users to review and approve each step the agent plans to execute before it proceeds. It ensures greater control, transparency and safety during the agent's decision-making process.

Inference with Human in the Loop

The Human-in-the-Loop functionality provides an interactive approach to multi-agent inference, where users maintain oversight throughout the execution process. This collaborative model bridges the gap between automated agent capabilities and human judgment.

Activation and Planning Process

Toggling on the Human-in-the-Loop button invokes the agent to provide a detailed plan about the steps it will perform to execute the tasks. This planning phase includes:

  • Task decomposition: Breaking down complex queries into manageable steps
  • Resource allocation: Identifying which agents or tools will be used for each step
  • Execution sequence: Determining the optimal order of operations
  • Risk assessment: Highlighting potential issues or dependencies

After the agent provides the comprehensive plan, it will ask for your approval or feedback before proceeding with execution.

User Interaction Controls

The interface provides intuitive controls for managing the agent's execution:

Approval Mechanism

The thumbs-up button serves as the approval mechanism. Clicking this button: - Approves the plan proposed by the agent - Signals the agent to proceed with executing your query - Initiates the execution phase with the current plan

Feedback and Iteration

The thumbs-down button enables the feedback loop. Clicking this button:

  • Prompts you to provide specific feedback on the generated plan
  • Allows you to highlight concerns, suggest modifications, or request clarifications
  • Triggers the agent to regenerate the plan incorporating your feedback
  • Maintains an iterative improvement cycle until approval is granted

Monitoring and Evaluation

You can view the critic score of the response by clicking the "Steps" dropdown. This feature provides:

  • Quality metrics: Numerical scores indicating the confidence level of each step
  • Performance indicators: Success probability assessments for different components
  • Transparency insights: Detailed breakdowns of the agent's reasoning process
  • Decision rationale: Explanations for why specific approaches were chosen

This comprehensive monitoring system ensures that users have full visibility into the agent's decision-making process and can make informed decisions about proceeding with the proposed plan.