Skip to main content

Memory and Orchestration

This section covers AgentMap's advanced features for building sophisticated, stateful workflows that can maintain context and intelligently route between different agents and capabilities.

Overview​

Memory and orchestration are key features that transform simple AgentMap workflows into intelligent, adaptive systems:

  • Memory Management: Enable agents to maintain conversation history and context
  • LangChain Integration: Leverage advanced memory strategies for optimal performance
  • Dynamic Orchestration: Route requests intelligently based on content and context
  • Prompt Management: Organize and reuse prompts across complex workflows

Quick Navigation​

Core Concepts​

Memory Systems​

AgentMap provides multiple memory approaches to suit different use cases:

  • Simple Memory: Basic conversation history tracking
  • LangChain Memory: Advanced strategies like buffer windows, token limits, and summarization
  • Shared Memory: Context sharing between multiple agents in a workflow

Orchestration Patterns​

Build intelligent routing systems that adapt to user input:

  • Dynamic Routing: Select the best agent based on request content
  • Multi-Strategy Selection: Combine algorithmic and AI-based routing
  • Context-Aware Decisions: Use conversation history to inform routing choices

Prompt Organization​

Maintain clean, reusable prompt libraries:

  • Registry-Based Prompts: Centralized prompt management
  • File-Based Prompts: Organize complex prompts in dedicated files
  • YAML Structures: Hierarchical prompt organization for complex scenarios

Integration Examples​

Memory + Orchestration​

Combine memory and orchestration for intelligent conversational flows:

graph_name,node_name,next_node,agent_type,input_fields,output_field,prompt,context
ChatBot,Router,,orchestrator,available_nodes|user_input|conversation_memory,next_node,"Route based on input and context","memory:{""type"":""buffer_window"",""k"":5}"
ChatBot,GeneralChat,,claude,user_input|conversation_memory,response,"I handle general conversation","memory:{""type"":""buffer_window"",""k"":5}"
ChatBot,TaskHelper,,claude,user_input|conversation_memory,response,"I help with specific tasks","memory:{""type"":""buffer_window"",""k"":5}"

Orchestration + Prompts​

Use managed prompts with intelligent routing:

graph_name,node_name,next_node,agent_type,input_fields,output_field,prompt,context
Support,Router,,orchestrator,available_nodes|user_input,next_node,prompt:router_instructions,
Support,TechSupport,,claude,user_input,response,prompt:technical_support,
Support,CustomerService,,claude,user_input,response,prompt:customer_service,

Complete Integration​

Combine all three features for maximum flexibility:

graph_name,node_name,next_node,agent_type,input_fields,output_field,prompt,context
Advanced,Router,,orchestrator,available_nodes|user_input|session_memory,next_node,yaml:workflows.yaml#routing.intelligent,"memory:{""type"":""summary"",""memory_key"":""session_memory""}"
Advanced,Specialist,,claude,user_input|session_memory,response,yaml:workflows.yaml#responses.specialist,"memory:{""type"":""summary"",""memory_key"":""session_memory""}"
Advanced,Generalist,,claude,user_input|session_memory,response,yaml:workflows.yaml#responses.generalist,"memory:{""type"":""summary"",""memory_key"":""session_memory""}"

Best Practices​

  1. Start Simple: Begin with basic memory and add complexity as needed
  2. Choose Appropriate Memory Types: Match memory strategy to conversation length and complexity
  3. Design Clear Node Descriptions: Help orchestrators make better routing decisions
  4. Organize Prompts Logically: Create maintainable prompt libraries
  5. Test Integration Points: Verify memory, routing, and prompts work together smoothly

Getting Started​

  1. Memory Management: Start with basic conversation memory
  2. LangChain Memory Integration: Explore advanced memory strategies
  3. Orchestration Patterns: Add intelligent routing to your workflows
  4. Prompt Management: Organize your prompts for reusability

Advanced Patterns​

Once you're comfortable with the basics, explore these advanced patterns:

  • Multi-Agent Conversations: Orchestrate between specialized agents while maintaining shared context
  • Hierarchical Routing: Use nested orchestrators for complex decision trees
  • Context-Aware Prompting: Adapt prompts based on conversation history and user context
  • Dynamic Memory Management: Adjust memory strategies based on conversation flow

These features work together to create sophisticated AgentMap workflows that can handle complex, stateful interactions while remaining maintainable and easy to understand.