MindLogic LogoMindLogic Docs
/
Dashboard

βš–οΈ Multi-Agent Debate System: Reaching Consensus via Logic Gates

When facing major and controversial business decisions, a single AI often hallucinates or carries bias. In these moments, we can introduce a Multi-Agent System.

Through MindLogic's visual graph, you can clearly see how various configured agents engage in a "Red vs. Blue" debate. Ultimately, through connection merging (AND/OR mechanisms), a Judge node distills the perfect decision.

Scenario Overview

In this workflow, let's assume we need to decide: "Should the company completely refactor the legacy monolithic architecture into microservices?"

We will build:

  1. Red Agent (Radical Tech Expert): Lists reasons why refactoring is a must.
  2. Blue Agent (Conservative CFO): Lists reasons to maintain the status quo and avoid risks.
  3. Judge (CEO Agent): Synthesizes both viewpoints and provides a compromised execution plan.

Node Orchestration Steps

Step 1: Create the Topic Node

Create an entity [Core Topic] and write in custom properties: issue = "Should we refactor the monolith to microservices?"

Step 2: Configure Red and Blue Agents

Create two parallel nodes: [Red Agent: CTO] and [Blue Agent: CFO]. Connect both from [Core Topic].

For the Red Node:

  1. Select the LLM plugin.
  2. System Prompt: You are a radical tech expert. Please list 3 hardcore technical reasons why refactoring is an absolute must.
  3. User Prompt: Topic: {{ node.inputs['issue'] }}
  4. Plugin output will be saved to node.outputs['red_argument'].

For the Blue Node:

  1. Select the LLM plugin.
  2. System Prompt: You are an extremely conservative CFO. Please list 3 fatal business risks and cost issues associated with refactoring.
  3. User Prompt: Topic: {{ node.inputs['issue'] }}
  4. Plugin output will be saved to node.outputs['blue_argument'].

Step 3: Merge Connections and Judge Node

Create a final node [The Judge: CEO]. Connect both the Red and Blue nodes simultaneously to the Judge node. By default, MindLogic sets connection logic to AND, meaning the Judge node won't trigger until both Red and Blue have finished outputting.

Configure the Judge Node:

  1. Select the LLM plugin.
  2. System Prompt: You are a wise CEO. Please synthesize both pro and con opinions, and provide a phased, risk-averse compromise plan.
  3. User Prompt: 
    Tech View: {{ node.inputs['red_argument'] }}
Finance View: {{ node.inputs['blue_argument'] }}

Results & Value

When you hit run, you will see the Red and Blue nodes execute concurrently in parallel! Once both complete, their results converge on the CEO node, generating a deep decision report addressing concerns from both sides.

The greatest charm of visual multi-agent orchestration is that the thinking process becomes transparent. Not only do you get the final outcome, but you can also trace the intermediate reasoning logic of both sides, ensuring every judgment is evidence-based.