Skip to content

feat: Multi-agent swimlane debugger with message flow tracing (CHI 2025) #193

@acailic

Description

@acailic

Paper Reference

  • Title: Interactive Debugging and Steering of Multi-Agent AI Systems
  • Authors: W Epperson, G Bansal, VC Dibia, A Fourney, et al.
  • Year: 2025
  • URL: https://dl.acm.org/doi/
  • Venue: ACM Conference on Human Factors in Computing Systems (CHI)

Paper Summary

Explores interfaces for debugging increasingly complex multi-agent AI systems with focus on effective support for debugging multi-agent teams. Identifies key challenges: coordination failures, emergent behavior, and communication breakdowns between agents.

Proposed Feature

Implement a multi-agent swimlane debugger:

Core Capabilities

  • Swimlane View: Display each agent as a horizontal lane with its actions plotted temporally
  • Message Flow Arrows: Show inter-agent communication as arrows between lanes
  • Coordination Analysis: Detect coordination failures, deadlocks, and communication gaps
  • Emergent Behavior Detection: Identify behaviors that emerge from agent interactions but aren't attributable to any single agent

Technical Approach

  • Add multi-agent session support to the SDK with per-agent event streams
  • Implement swimlane visualization component with D3.js
  • Add inter-agent message tracking and visualization
  • Build coordination analysis algorithms

Impact

Essential for users debugging multi-agent systems (CrewAI, AutoGen, etc.). Currently no open-source debugger provides this capability.

Labels

enhancement, paper-inspired, frontend, multi-agent

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions