Agentic System Architecture

A visual map showing how LLMs, agent components, agent types, and workflow patterns interconnect to create intelligent AI systems

Foundation (LLM)
Agent Components
Agent Types
Workflow Patterns
💡 How to Use This Map
Hover over any node to see detailed descriptions. The architecture flows from center outward: LLM → Components → Types → Patterns
🧠
LLM
🎭
Persona
Identity & Role
📚
Knowledge
Information Access
💬
Prompting
Communication Strategy
🔧
Tools
Execution Capabilities
🔄
Interaction
Input/Output
➡️
Direct
Simple Relay
Augmented
Enhanced Context
Dynamic
Real-time Adaptation
🤖
Autonomous
Full Reasoning
🔗
Chaining
Sequential Steps
🔀
Routing
Smart Direction
Parallelization
Concurrent Work
🔄
Evaluator
Iterative Refinement
🎯
Orchestrator
Dynamic Planning
🧠 Foundation: LLM
The Large Language Model serves as the "brain" of every agent, providing core knowledge, reasoning capabilities, and natural language understanding. All agent capabilities flow from this foundation.
🔵 Agent Components
The 5 building blocks that define how an agent operates:
  • Persona: Identity, role, tone, boundaries
  • Knowledge: Training data, fine-tuning, tools, memory
  • Prompting: Zero-shot, few-shot, Chain-of-Thought
  • Tools: APIs, databases, external systems
  • Interaction: How input/output is handled
🟢 Agent Types
Agents exist on a spectrum of sophistication:
  • Direct: Simple relay to LLM
  • Augmented: Adds context before LLM call
  • Dynamic: Adapts context in real-time
  • Autonomous: Full planning & decision-making
🟡 Workflow Patterns
How multiple agents coordinate to solve problems:
  • Chaining: Sequential dependent steps
  • Routing: Direct to specialized agents
  • Parallelization: Concurrent execution
  • Evaluator-Optimizer: Iterative refinement
  • Orchestrator-Workers: Dynamic coordination