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 (16+ Built-in Types)

Core Agents (10 types): Default, Echo, Branching, Success, Failure, Input, Graph, Human, Orchestrator, Summary
LLM Agents (4 types): OpenAI (GPT), Anthropic (Claude), Google (Gemini), LLM (base) with unified interface
Storage Agents (6 types): CSV (reader/writer), JSON (reader/writer), File (reader/writer), Vector (reader/writer), Document (reader/writer), Blob Storage
Custom Agent Support: Full scaffolding system for extension with service-aware generation

🤖 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 (run)
Advanced scaffolding commands with service integration (scaffold)
Bundle management and caching (update-bundle)
Configuration management and comprehensive validation (init-config, validate, diagnose, refresh)

Available Commands: run, scaffold, update-bundle, init-config, validate, diagnose, refresh

Repository Workflows - Execute workflows from CSV repository:

# Direct repository execution
agentmap run workflow/GraphName

# Repository with CSV file  
agentmap run workflows/hello_world.csv --graph HelloWorld

# Traditional file execution
agentmap run path/to/workflow.csv --graph GraphName

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

High-Performance Bundle System

Bundle-based caching: Intelligent graph compilation and caching for 10x faster execution
Static analysis optimization: Declaration-based analysis eliminates runtime overhead
Repository workflows: Execute workflows directly with agentmap run workflow/GraphName syntax
Composite key lookups: Efficient bundle retrieval using CSV hash + graph name
Smart invalidation: Bundles automatically refresh when CSV content changes

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 with unified service registry

Bundle-Based Execution

Pre-compiled workflows: CSV files compiled to cached bundle objects for faster execution
Static analysis: Declaration-based graph validation eliminates runtime overhead
Intelligent caching: Bundles automatically invalidate when source CSV changes
Repository integration: Direct execution from workflow repositories without local files

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
`GraphName`	✅	Workflow identifier	`ChatBot`, `DocumentProcessor`
`Node`	✅	Unique node name within graph	`GetInput`, `ProcessData`, `SaveResults`
`Edge`	❌	Direct connection to next node	`NextNode`, `func:custom_router`
`Context`	❌	Node configuration (JSON or text)	`{"memory_key":"chat_history"}`
`AgentType`	❌	Type of agent to use	`openai`, `claude`, `csv_reader`
`Success_Next`	❌	Next node on success	`ProcessData`, `Success\|Backup`
`Failure_Next`	❌	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"`

Note: Column names support case-insensitive matching and aliases. For example, GraphName accepts graph_name, workflow_name; AgentType accepts agent_type, Agent; Node accepts node_name, NodeName, etc.

Advanced Routing Patterns

Conditional Branching
Multiple Targets
Function-Based Routing

Conditional Branching Example

⚠️ Validation Results

2 errors

Row 1, graph_name:Required header 'graph_name' is missing

Row 1, next_node:Required header 'next_node' is missing

GraphName	Node	Edge	Context	AgentType	Success_Next	Failure_Next	Input_Fields	Output_Field	Prompt
DataFlow	Validate		Conditional validation logic	branching	Transform	ErrorHandler	raw_data	validation_result

10 columns, 1 rows

Parallel Processing Example

⚠️ Validation Results

2 errors

Row 1, graph_name:Required header 'graph_name' is missing

Row 1, next_node:Required header 'next_node' is missing

GraphName	Node	Edge	Context	AgentType	Success_Next	Failure_Next	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 Example

⚠️ Validation Results

2 errors

Row 1, graph_name:Required header 'graph_name' is missing

Row 1, next_node:Required header 'next_node' is missing

GraphName	Node	Edge	Context	AgentType	Success_Next	Failure_Next	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
`human`	Human-in-the-loop	Any fields	Prompts human for 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
`graph`	Sub-graph execution	Any fields	Executes nested graph workflow
`orchestrator`	Intelligent routing	Query and context	Routes to best available agent
`summary`	Content summarization	Text content	Returns summarized content

LLM Agent Types

Agent Type	Provider	Features	Configuration
`llm`	Generic	Base LLM agent, unified interface	Provider, model, temperature, memory settings
`openai` (aliases: `gpt`)	OpenAI	GPT models, memory	Model, temperature, memory settings
`anthropic` (alias: `claude`)	Anthropic	Claude models, memory	Model, temperature, memory settings
`google` (alias: `gemini`)	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
`document_reader`	Read documents	`collection` (file path)	Document content with metadata
`document_writer`	Write documents	`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

Feature Status & Availability

Implementation Status

🟢 Stable: Core agents (10), LLM integration (4 providers), storage services (6 types), scaffolding system
🟡 Beta: Vector database integration, orchestrator routing, bundle caching system
🔵 Validated: All documented CLI commands and CSV schema verified against current implementation (v2024.09.03)

CLI Commands Status

Available Commands (verified in main_cli.py):

run - Execute workflows with bundle caching and repository support
scaffold - Generate agents with service integration
update-bundle - Manage bundle cache and compilation
validate - Validate CSV and configurations
diagnose - System diagnostics and health checks
refresh - Refresh caches and registries
init-config - Initialize configuration files

Note: Previously documented commands export, resume, validate-csv, validate-config, and validate-all are not currently implemented.

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

Architecture Guide: Deep dive into AgentMap's service-oriented architecture, bundle system, and dependency injection
CSV Schema Reference: Complete schema documentation with column specifications
Agent Types Reference: All available agent types with implementation details
CLI Commands: Command-line tools and options

🎯 Core Agentic AI Features​

Autonomous Multi-Agent Workflows​

RAG AI & Vector Database Integration​

Agent Ecosystem (16+ Built-in Types)​

🤖 AI & LLM Capabilities​

Multi-LLM Integration​

Memory Management System​

Advanced AI Features​

💾 Storage & Integration​

Universal Storage Support​

Storage Configuration​

🛠️ Developer Experience​

Service-Aware Scaffolding System​

Powerful CLI System​

Advanced Code Generation​

Development Tools​

📊 Execution & Monitoring​

High-Performance Bundle System​

Execution Tracking System​

Success Policies​

State Management​

🏗️ Architecture & Extensibility​

Service-Oriented Design​

Bundle-Based Execution​

Advanced Routing​

CSV Schema System​

Core Columns​

Advanced Routing Patterns​

Conditional Branching Example

Parallel Processing Example

Function-Based Routing Example

Agent Types Reference​

Core Agent Types​

LLM Agent Types​

Storage Agent Types​

File Operations​

Structured Data​

Cloud Storage​

Vector Databases​

Feature Status & Availability​

Implementation Status​

CLI Commands Status​

Next Steps​

Related Documentation​

🎯 Core Agentic AI Features

Autonomous Multi-Agent Workflows

RAG AI & Vector Database Integration

Agent Ecosystem (16+ Built-in Types)

🤖 AI & LLM Capabilities

Multi-LLM Integration

Memory Management System

Advanced AI Features

💾 Storage & Integration

Universal Storage Support

Storage Configuration

🛠️ Developer Experience

Service-Aware Scaffolding System

Powerful CLI System

Advanced Code Generation

Development Tools

📊 Execution & Monitoring

High-Performance Bundle System

Execution Tracking System

Success Policies

State Management

🏗️ Architecture & Extensibility

Service-Oriented Design

Bundle-Based Execution

Advanced Routing

CSV Schema System

Core Columns

Advanced Routing Patterns

Agent Types Reference

Core Agent Types

LLM Agent Types

Storage Agent Types

File Operations

Structured Data

Cloud Storage

Vector Databases

Feature Status & Availability

Implementation Status

CLI Commands Status

Next Steps

Related Documentation