Hi,
I thought I will be able to create a multi agent system using Gale. But it looks like it only allows creation of agents. Is there a way to create a multi agent system similar to how we can do through langgraph. Also how is shared persistent state and memory managed through gale.
Hi Rajib,
Great question! You’re absolutely right that the platform is designed for multi-agent orchestration, similar to LangGraph patterns. Here’s how it works:
Multi-Agent Architecture The Kore.ai Agent Platform is specifically built for multi-agent systems using the Supervisor Pattern. You build individual agents focused on specific domains (like HR policies or sales data), then use a Supervisorto coordinate them into a unified system.
How the Supervisor Pattern Works:
- Build Specialized Agents - Create domain-specific agents with deep expertise
- Deploy the Supervisor - Acts as the central orchestrator that manages and coordinates multiple AI agents
- Intelligent Coordination - The supervisor breaks down complex requests into smaller subtasks, delegates them to appropriate specialized agents, then aggregates results into a unified response
The supervisor handles task decomposition, agent selection, parallel execution coordination, conflict resolution, and response synthesis—exactly the kind of orchestration you’d see in LangGraph.
Supporting Documentation:
- Supervisor Pattern Documentation - Complete architectural approach and workflow details
- Multi-Agent System Overview - Core concepts and advantages over single-agent systems
- Multi-Agent Orchestration - Platform capabilities for agent coordination
- Agent Platform Features - Advanced orchestration capabilities
So yes, you can definitely create the LangGraph-style multi-agent system you’re looking for using our Supervisor Pattern!
Hope this helps clarify the approach.
Thanks Scott, so it will be a LLM that will always determine the orchestration. Is there a way to do control flow engineering more like a deterministic workflow type of approach where I as a human decide the flow in a pre-defined graph