AG // agentology
🧠 Anatomy

Agent Architecture Patterns

Visual overview of common agent architectural patterns and their flow structures

#architecture #patterns #design

This entry demonstrates visual architectural patterns using Mermaid diagrams instead of ASCII art.

The Basic Agent Loop

graph TD
  A[OBSERVE<br/>read_input<br/>get_state<br/>parse_env] --> B[REASON<br/>think<br/>plan<br/>decide]
  B --> C[ACT<br/>call_tools<br/>output]
  C --> D[OBSERVE<br/>results]
  D -.repeat.-> A

  style A fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc
  style B fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc
  style C fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc
  style D fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc
observe → reason → act → repeat

The fundamental cycle underlying all agent behavior.

ReAct Pattern

graph LR
  T1[Thought:<br/>need population data] --> A1[Action:<br/>search tokyo_population]
  A1 --> O1[Observation:<br/>13.96M residents]
  O1 --> T2[Thought:<br/>have answer now]
  T2 --> F[finish]

  style T1 fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc
  style A1 fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc
  style O1 fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc
  style T2 fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc
  style F fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc
thought → action → observation loop

Explicit reasoning traces interleaved with actions.

Hierarchical Agent System

graph TD
  M[Manager Agent<br/>strategic planning] --> S1[Specialist Agent 1<br/>code generation]
  M --> S2[Specialist Agent 2<br/>testing]
  M --> S3[Specialist Agent 3<br/>documentation]

  S1 --> T1[Tool: IDE]
  S2 --> T2[Tool: Test Runner]
  S3 --> T3[Tool: Doc Generator]

  style M fill:#0a0a0a,stroke:#00ff00,stroke-width:2px,color:#cccccc
  style S1 fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc
  style S2 fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc
  style S3 fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc
  style T1 fill:#1a1a1a,stroke:#333333,stroke-width:1px,color:#666666
  style T2 fill:#1a1a1a,stroke:#333333,stroke-width:1px,color:#666666
  style T3 fill:#1a1a1a,stroke:#333333,stroke-width:1px,color:#666666
multi-level agent coordination

Coordinated multi-agent systems with specialized roles.

Memory Architecture

graph TB
  Input[User Input] --> STM[Short-Term Memory<br/>conversation context]
  STM --> Proc[Processing<br/>LLM reasoning]
  Proc --> LTM[Long-Term Memory<br/>vector store]
  LTM -.retrieval.-> Proc
  Proc --> Output[Agent Output]

  LTM --> EXT[External Memory<br/>databases / files]

  style Input fill:#0a0a0a,stroke:#333333,stroke-width:1px,color:#cccccc
  style STM fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc
  style Proc fill:#0a0a0a,stroke:#00ff00,stroke-width:2px,color:#cccccc
  style LTM fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc
  style EXT fill:#0a0a0a,stroke:#333333,stroke-width:1px,color:#666666
  style Output fill:#0a0a0a,stroke:#333333,stroke-width:1px,color:#cccccc
agent memory systems

Memory layers enabling persistent agent knowledge.

State Machine Pattern

stateDiagram-v2
  [*] --> Idle
  Idle --> Planning: task_received
  Planning --> Executing: plan_ready
  Executing --> Observing: action_complete
  Observing --> Planning: need_replan
  Observing --> Executing: continue_plan
  Observing --> Complete: goal_achieved
  Complete --> [*]

  Executing --> Error: failure
  Error --> Planning: retry
  Error --> [*]: abort
agent state transitions

State-based agent execution model.

Tool Use Flow

sequenceDiagram
  participant Agent
  participant Executor
  participant Tool
  participant Environment

  Agent->>Executor: call_tool(search, "AI agents")
  Executor->>Tool: invoke(search)
  Tool->>Environment: web_request
  Environment-->>Tool: results
  Tool-->>Executor: formatted_response
  Executor-->>Agent: observation
  Agent->>Agent: reason_about_results
agent-tool interaction

Sequence of tool invocation and result processing.

Advantages

  • clarity: visual representation more intuitive than text descriptions
  • maintainability: diagram source code stored as text in markdown
  • consistency: unified styling across all diagrams
  • scalability: complex architectures represented concisely

Usage

Replace ASCII art with Mermaid syntax:

<Diagram type="mermaid" caption="your caption">
{\`graph TD
    A[Node A] --> B[Node B]
\`}
</Diagram>

Supports flowcharts, sequence diagrams, state diagrams, and more.