Multi-Agent Characteristics

These dimensions control how agents interact with each other. They determine who can talk to whom, how information flows, and whether agents can collaborate on decisions.

Network Topology

This dimension controls how agents are connected and which coordination patterns are possible. The topology shapes what information can flow between agents and how quickly coordination happens.

How frameworks handle this

Graph-based frameworks let you define explicit topologies as directed graphs at design time. You can create hierarchical, peer-to-peer, or cyclic structures, but the topology stays fixed during execution.

Role-based frameworks create topology implicitly through task decomposition and role assignments. This typically results in sequential pipelines or manager-worker hierarchies with limited lateral interaction.

GABM frameworks centralize interaction through an environment controller (Game Master), enforcing a star topology where agents only interact with the environment, not directly with each other.

Network Topology Examples

Graph-based: Supervisor

S A1 A2 a1 a2 a3 a4

Hierarchical tree structure

Graph-based: Network

A1 A2 A3 A4

Fully connected graph

Role-based: Task hierarchy

Orch Agent Agent Agent Task Task

Orchestrator → Agents → Tasks

GABM: Star topology

GM a a a a

All agents connect only to Game Master (GM)

Framework Topology
Graph-based Flexible; fixed once defined
Role-based Task hierarchy or sequence
GABM Fixed star; GM at center

Communication Pattern

This dimension controls how information moves between agents. Communication can happen through state propagation, message passing, or environment observation.

How frameworks handle this

Graph-based frameworks realize communication through deterministic state propagation along predefined edges in the execution graph, rather than explicit message passing between agents.

Role-based frameworks route communication primarily through task assignment, intermediate result passing, and reporting mechanisms embedded in hierarchical or sequential coordination structures.

GABM frameworks eliminate direct inter-agent messaging. All interaction is mediated through a shared environment that aggregates agent actions and produces new observations for subsequent reasoning.

Framework Communication
Graph-based Shared state along edges
Role-based Task-mediated; limited peer querying
GABM Only via GM; no peer-to-peer

Collaboration

This dimension captures whether frameworks provide explicit mechanisms for agents to jointly coordinate decisions beyond procedural execution.

How frameworks handle this

Graph-based frameworks do not support first-class collaboration. They rely entirely on procedural coordination enforced by predefined workflows.

Role-based frameworks enable limited forms of collaboration through delegation, task handoff, and manager-worker coordination among role-specialized agents. However, they generally lack support for multi-turn negotiation or collective decision-making protocols.

GABM frameworks expose no explicit collaboration mechanisms. Collective behavior emerges indirectly from shared environment dynamics rather than from intentional or negotiated coordination among agents.

Framework Collaboration
Graph-based None; procedural only
Role-based Partial; delegation-based
GABM None; emergent only