AgentMap Core Features & Capabilities

AgentMap is a sophisticated agentic AI orchestration framework that transforms simple CSV files into powerful autonomous multi-agent AI systems. This comprehensive guide covers the complete feature set for building RAG AI applications, multi-agent workflows, and LLM orchestration systems.

🎯 Core Agentic AI Features

Autonomous Multi-Agent Workflows

• Agentic decision-making with intelligent routing and autonomous behavior
• Multi-agent collaboration where specialized agents coordinate and communicate
• Self-directed execution with agents that adapt and respond to changing conditions
• Hierarchical agent systems with supervisor and worker agent patterns
• Event-driven autonomy where agents react intelligently to triggers and state changes

RAG AI & Vector Database Integration

• Native vector database support (Chroma, FAISS, Pinecone) for retrieval-augmented generation
• Semantic search agents that intelligently query knowledge bases
• Document processing pipelines with chunking, embedding, and retrieval
• Knowledge-aware LLM agents that combine reasoning with retrieved context
• Multi-modal RAG systems supporting text, code, and structured data

Agent Ecosystem (20+ Built-in Types)

• Core Agents: Default, Echo, Branching, Success/Failure, Input
• LLM Agents: OpenAI (GPT), Anthropic (Claude), Google (Gemini) with unified interface
• Storage Agents: CSV, JSON, File operations with local and cloud support
• Advanced Agents: Vector databases, Orchestrator, Summary, Graph agents
• Custom Agent Support: Full scaffolding system for extension

🤖 AI & LLM Capabilities

Multi-LLM Integration

• Unified interface across OpenAI, Anthropic, Google providers
• Configurable models, temperature, and parameters per node
• Automatic prompt template processing with field substitution
• Memory management with conversation history and context retention

Memory Management System

• Multiple memory types: Buffer, Buffer Window, Summary, Token Buffer
• Declarative memory configuration through CSV Context field
• Automatic serialization/deserialization between nodes
• Shared memory across multi-agent workflows

Advanced AI Features

• Intelligent orchestration with dynamic routing based on content analysis
• Vector database integration for semantic search and document retrieval
• Document processing with chunking and metadata extraction
• Prompt management system with registry, file, and YAML references

💾 Storage & Integration

Universal Storage Support

• Local Storage: CSV, JSON, file operations with LangChain integration
• Cloud Storage: Azure Blob, AWS S3, Google Cloud Storage with URI-based access
• Databases: Firebase integration, vector stores (Chroma, FAISS)
• Document Processing: PDF, Word, Markdown, HTML with intelligent chunking

Storage Configuration

• Centralized storage configuration with provider-specific settings
• Environment variable support for credentials
• Container/bucket mapping with logical names
• Multiple authentication methods per provider

🛠️ Developer Experience

Service-Aware Scaffolding System

AgentMap's most powerful productivity feature - an intelligent code generation system that analyzes CSV workflows and automatically creates service-integrated agent classes:

• Service-aware code generation: Automatically detects service requirements from CSV context and generates agents with proper LLM, storage, vector, and memory service integration
• Multi-architecture support: Unified storage vs. separate service protocols based on requirements analysis
• Template system: Sophisticated IndentedTemplateComposer with modular agent and function templates
• Agent registry integration: Conflict detection to avoid scaffolding existing agents
• Complete workflow integration: Scaffold → customize → test → deploy development cycle

Supported Services: LLM (OpenAI, Anthropic, Google), Storage (CSV, JSON, File, Vector, Memory), Node Registry

Example Usage:

# Service-aware scaffolding with automatic service detection
agentmap scaffold --graph IntelligentWorkflow

# Generated agents with service integration
class DataAnalyzerAgent(BaseAgent, LLMCapableAgent, StorageCapableAgent):
    # Automatic service injection and usage examples included

Powerful CLI System

• Workflow execution with state management and real-time feedback
• Advanced scaffolding commands with service integration
• Graph compilation and export capabilities
• Interactive configuration management and comprehensive validation

Advanced Code Generation

• Context-aware templates: Service requirements parsed from CSV context fields
• Protocol integration: Automatic inheritance from LLMCapableAgent, StorageCapableAgent, etc.
• Usage examples: Generated code includes service integration examples and best practices
• Function scaffolding: Complete routing function generation with context-aware logic

Development Tools

• Hot reloading for rapid development cycles
• Comprehensive logging and debugging support
• Execution tracking with configurable success policies
• Performance monitoring and metrics

📊 Execution & Monitoring

Execution Tracking System

• Two-tier tracking: Minimal (always on) and Detailed (optional)
• Policy-based success evaluation with multiple strategies
• Real-time execution path monitoring
• Performance metrics and timing information

Success Policies

• All nodes must succeed
• Final node success only
• Critical nodes success
• Custom policy functions

State Management

• Immutable state transitions with comprehensive data flow
• Multiple state formats support (dict, Pydantic, custom)
• Memory serialization and field mapping
• Error handling and recovery mechanisms

🏗️ Architecture & Extensibility

Service-Oriented Design

• Clean separation of concerns with dependency injection
• Pluggable architecture with consistent interfaces
• Agent contract system for custom implementations
• Storage abstraction layers

Advanced Routing

• Conditional branching based on execution success
• Function-based routing with custom logic
• Multi-target routing for parallel processing
• Orchestrator-based intelligent routing

CSV Schema System

Core Columns

Column	Required	Description	Examples
graph_name	✅	Workflow identifier	ChatBot, DocumentProcessor
Node	✅	Unique node name within graph	GetInput, ProcessData, SaveResults
next_node	❌	Direct connection to next node	NextNode, func:custom_router
Context	❌	Node configuration (JSON or text)	{"memory_key":"chat_history"}
agent_type	❌	Type of agent to use	openai, claude, csv_reader
next_on_success	❌	Next node on success	ProcessData, Success|Backup
next_on_failure	❌	Next node on failure	ErrorHandler, Retry
input_fields	❌	State fields to extract as input	user_input|context|memory
output_field	❌	Field to store agent output	response, processed_data
Prompt	❌	Agent prompt or configuration	"You are helpful: {input}", prompt:system_instructions
Description	❌	Documentation for the node	"Validates user input format"

Advanced Routing Patterns

Conditional Branching

Conditional Branching Example

graph_name	node_name	next_node	context	agent_type	next_on_success	next_on_failure	input_fields	output_field	prompt
DataFlow	Validate		Conditional validation logic	branching	Transform	ErrorHandler	raw_data	validation_result

10 columns, 1 rows

Multiple Targets

Parallel Processing Example

graph_name	node_name	next_node	context	agent_type	next_on_success	next_on_failure	input_fields	output_field	prompt
Parallel	Distribute		Split data for parallel processing	default	ProcessA|ProcessB|ProcessC		data	distributed_tasks

10 columns, 1 rows

Function-Based Routing

Function-Based Routing Example

graph_name	node_name	next_node	context	agent_type	next_on_success	next_on_failure	input_fields	output_field	prompt
Smart	Classifier	func:choose_specialist	Route to specialized agents	default			user_query	classification

10 columns, 1 rows

Agent Types Reference

Core Agent Types

Agent Type	Purpose	Input Behavior	Output Behavior
default	Basic processing	Any fields	Returns message with prompt
echo	Pass-through	First input field	Returns input unchanged
input	User interaction	Ignored	Prompts user, returns input
branching	Conditional routing	Looks for success indicators	Returns routing decision
success	Always succeeds	Any	Returns success message
failure	Always fails	Any	Returns failure message

LLM Agent Types

Agent Type	Provider	Features	Configuration
openai (aliases: gpt, chatgpt)	OpenAI	GPT models, memory	Model, temperature, memory settings
claude (alias: anthropic)	Anthropic	Claude models, memory	Model, temperature, memory settings
gemini (alias: google)	Google	Gemini models, memory	Model, temperature, memory settings

Storage Agent Types

File Operations

Agent Type	Purpose	Required Input	Output
file_reader	Read documents	collection (file path)	Document content with metadata
file_writer	Write files	collection (path), data	Operation result

Structured Data

Agent Type	Purpose	Required Input	Output
csv_reader	Read CSV files	collection (file path)	Parsed CSV data
csv_writer	Write CSV files	collection (path), data	Operation result
json_reader	Read JSON files	collection (file path)	JSON data
json_writer	Write JSON files	collection (path), data	Operation result

Cloud Storage

Agent Type	Purpose	URI Format	Authentication
cloud_json_reader	Read from cloud	azure://container/file.json	Connection string/keys
cloud_json_writer	Write to cloud	s3://bucket/file.json	AWS credentials

Vector Databases

Agent Type	Purpose	Configuration	Use Cases
vector_reader	Similarity search	Store configuration	Document retrieval, semantic search
vector_writer	Store embeddings	Store configuration	Knowledge base building

Next Steps

Ready to Get Started?

• Quick Start: Begin with our 5-minute tutorial
• Build Your First Workflow: Follow the quick start guide
• Explore Examples: Check out example workflows

Deep Dive Topics

• State Management: Understand data flow between agents
• Agent Development: Create custom agents
• Infrastructure Guide: Work with files, databases and cloud storage

Related Documentation

• CSV Schema Reference: Complete schema documentation
• Agent Types Reference: All available agent types
• CLI Commands: Command-line tools and options