Roundtable icon

Roundtable

Orchestrates multi-agent LLM discussions with a single model

Roundtable

AI-generated

Overview

This node orchestrates multi-agent discussions using a single large language model (LLM) by simulating multiple participants with distinct roles and system prompts. It is useful for scenarios where collaborative brainstorming, expert panel discussions, or multi-perspective analysis is needed, such as designing a new product feature or strategizing business plans. The node facilitates a structured conversation with a host (facilitator) and multiple participants, running a specified number of discussion rounds and producing a formatted output document summarizing the discussion or providing raw conversation logs.

Use Case Examples

  1. A product team simulating a discussion between a Product Manager, Engineer, and Designer to evaluate a new app feature.
  2. A business strategy session with multiple expert personas discussing market entry strategies.
  3. A creative brainstorming session generating a creative brief from diverse expert inputs.

Properties

Name Meaning
LLM Connection Notice to connect an LLM node to simulate multiple participants using different system prompts.
Discussion Topic The main topic to be discussed, e.g., 'Design a new mobile app feature'.
Context Additional background information or constraints relevant to the discussion.
Speaking Points Specific points to address in the discussion, entered one per line.
Host Configuration Settings for the discussion facilitator, including role name, system prompt, and response randomness (temperature).
Participants List of participants with their roles, system prompts, and response randomness (temperature). At least two participants are required.
Number of Rounds Number of discussion rounds to simulate (1-20).
Output Format Format of the output document summarizing the discussion.
Advanced Options Additional options such as enabling verbose logging of the conversation.

Output

JSON

  • topic - The main discussion topic provided as input.
  • output - The formatted output document or raw conversation based on the selected output format.
  • format - The chosen output format type.
  • participants - Array of participant objects with their IDs and roles.
  • messageCount - The total number of messages exchanged in the conversation log.
  • iterations - The number of discussion rounds executed.
  • conversationLog - (Optional) Detailed conversation log if verbose logging is enabled.

Dependencies

  • Requires connection to an LLM node to simulate participants using the same model with different system prompts.

Troubleshooting

  • Error if no LLM node is connected: Ensure an LLM node is connected to the input to simulate participants.
  • Error if fewer than two participants are defined: Define at least two participants with roles and system prompts to enable a meaningful discussion.
  • If the node fails unexpectedly, check the input parameters for completeness and correctness, especially the topic, participants, and host configuration.
  • Enable verbose logging in advanced options to get detailed conversation logs for debugging.

Discussion