Skip to main content

Welcome to AgentMap

Build Multi-Agent AI Workflows with Simple CSV Files

AgentMap allows you to create sophisticated multi-agent AI systems using familiar CSV configuration files. Define workflows where AI agents collaborate autonomously to solve complex problems.

What is AgentMap?​

AgentMap is a multi-agent orchestration framework that enables you to:

  • Define workflows in CSV - Use simple, version-controllable CSV files instead of complex YAML
  • Orchestrate AI agents - Combine LLM reasoning, data processing, and custom logic seamlessly
  • Build autonomous systems - Agents make decisions and route intelligently based on context
  • Scale production workloads - Built-in monitoring, error handling, and performance tracking

Core Concepts​

Agents​

All agents inherit from BaseAgent and implement a process() method:

class WeatherAgent(BaseAgent):
    def process(self, inputs: Dict[str, Any]) -> Any:
        location = inputs.get('location', 'Unknown')
        # Call weather API and return data
        return f"Weather in {location}: Sunny, 72°F"

Built-in Agent Types: LLM agents, file operations, data processing, routing, and more.

Services​

Infrastructure services are injected via protocols for clean architecture:

# Protocol-based injection for business services
class LLMAgent(BaseAgent, LLMCapableAgent):
    def configure_llm_service(self, llm_service: LLMServiceProtocol):
        self._llm_service = llm_service

Service Categories: LLM services, storage services, execution tracking, state management.

Workflows​

CSV files define the workflow structure - agents, connections, and data flow:

workflow,node,description,type,next_node,error_node,input_fields,output_field,prompt
SimpleBot,GetInput,Get user question,input,ProcessQuestion,End,,question,Enter your question:
SimpleBot,ProcessQuestion,Process with AI,llm,GetInput,Error,question,response,Answer this question: {question}
SimpleBot,Error,Handle errors,echo,End,,error,error_msg,Sorry there was an error
SimpleBot,End,Complete workflow,echo,,,response,final_output,{response}

Key Elements: Node definitions, data routing, error handling, state management.

Quick Example​

Here's a simple 3-agent workflow for document analysis:

workflow,node,description,type,next_node,error_node,input_fields,output_field,prompt
DocAnalyzer,LoadDoc,Load document,file_reader,AnalyzeContent,Error,file_path,document,
DocAnalyzer,AnalyzeContent,Extract insights,llm,CreateSummary,Error,document,insights,Analyze this document and extract key insights: {document}
DocAnalyzer,CreateSummary,Create summary,llm,End,Error,insights,summary,Create an executive summary from these insights: {insights}
DocAnalyzer,Error,Handle errors,echo,End,,error,error_msg,Analysis failed
DocAnalyzer,End,Complete analysis,echo,,,summary,final_result,{summary}

What happens:

  1. LoadDoc agent reads a document file
  2. AnalyzeContent agent uses LLM to extract insights
  3. CreateSummary agent creates an executive summary
  4. Data flows automatically between agents via input_fields and output_field

Documentation Overview​

🏃‍♂️ Getting Started​

Build your first workflow in 5 minutes with step-by-step guidance.

🎓 Learning Paths​

Progressive learning guides that build from basic concepts to advanced patterns:

📚 Learning Guides​

Progressive lessons with downloadable examples:

📖 Guides​

In-depth development and deployment guides:

  • Development - Agent creation, testing, best practices
  • Deployment - Production deployment and monitoring

📚 Reference​

Complete specifications and API documentation:

  • Agents - Built-in agent types and development patterns
  • Services - Service architecture and protocols
  • CSV Schema - Complete workflow definition format

🤝 Contributing​

Architecture guides and contribution patterns:

Ready to start building?

Begin Quick Start Guide →