Agentic Design Patterns

A collection of examples demonstrating various agentic AI design patterns implemented across multiple frameworks including LangChain, Google ADK, and CrewAI.

Patterns Covered

1. Prompt Chaining

• 1_prompt_chaining_langchain.py - Sequential prompt execution where output of one prompt feeds into the next

2. Routing

• 2_1_routing_langchain.py - LangChain implementation of request routing
• 2_2_routing_google_adk.py - Google ADK implementation using sub-agents

3. Parallelization

• 3_1_parallelization_langchain.py - LangChain parallel task execution with result synthesis
• 3_2_parallelization_google_adk.py - Google ADK ParallelAgent for concurrent research

4. Self-Reflection

• 4_1_self_reflection_langchain.py - Iterative code generation with AI critique
• 4_2_self_reflection_google_adk.py - Draft generation with fact-checking pipeline

5. Tool/Function Calling

• 5_1_tool_function_call_langchain.py - LangChain agent with custom tools
• 5_2_tool_function_call_crewai.py - CrewAI agent with stock price lookup tool
• 5_3_tool_function_call_google_adk.py - Google ADK with Google Search tool
• 5_4_tool_function_call_google_adk.py - Code execution with BuiltInCodeExecutor
• 5_5_tool_function_call_google_adk.py - Vertex AI Search integration

6. Planning

• 6_1_planning_crewai.py - Task planning and sequential execution with CrewAI

7. Multi-Agent Collaboration

• 7_1_multi_agent_collaboration_crewai.py - Sequential multi-agent workflow (researcher → writer) for blog content creation
• 7_21_multi_agent_collaboration_google_adk.py - Custom BaseAgent implementation with parent-child hierarchy and TaskExecutor
• 7_22_multi_agent_collaboration_google_adk.py - LoopAgent with ConditionChecker for iterative workflow automation
• 7_23_multi_agent_collaboration_google_adk.py - SequentialAgent pipeline for data fetch and process workflow
• 7_24_multi_agent_collaboration_google_adk.py - ParallelAgent for concurrent weather and news data fetching
• 7_25_multi_agent_collaboration_google_adk.py - AgentTool wrapper for nested agent delegation (Artist → ImageGen)

8. Memory Management

• 8_11_memory_management_google_adk.py - Google ADK session state management with output_key
• 8_12_memory_management_google_adk.py - Google ADK ToolContext for state management within tools
• 8_21_memory_management_langchain.py - LangChain ConversationBufferMemory with LLMChain for travel agent
• 8_22_memory_management_langchain.py - LangChain ConversationBufferMemory with ChatPromptTemplate for chat models
• 8_31_memory_management_langgraph.py - LangGraph InMemoryStore with semantic search and embeddings

10. Model Context Protocol (MCP)

• 10_1_model_context_protocol_google_adk.py - Google ADK MCP toolset with filesystem server (npx)
• 10_21_mcp_server_fastmcp.py - FastMCP server implementation with greet tool
• 10_22_mcp_client_google_adk.py - Google ADK client connecting to FastMCP server via HTTP

11. Goal Setting and Monitoring

• 11_1_goal_setting_and_monitoring_langchain.py - LangChain agent with iterative code generation and goal evaluation

12. Exception Handling

• 12_1_exception_handling_google_adk.py - Google ADK SequentialAgent with primary/fallback handlers for robust error handling

13. Human in the Loop

• 13_1_human_in_the_loop_google_adk.py - Google ADK with tool callbacks for personalization and human escalation

14. Retrieval-Augmented Generation (RAG)

• 14_11_rag_google_search_google_adk.py - Google ADK agent with Google Search tool for RAG
• 14_12_rag_vector_ai_google_adk.py - Google ADK VertexAiRagMemoryService with vector similarity search
• 14_21_rag_langchain.py - LangChain RAG with Weaviate vector store and OpenAI embeddings

15. Agent-to-Agent (A2A) Protocol

• 15_11_agent_to_agent_a2a_agent_google_adk.py - Google ADK calendar agent with CalendarToolset
• 15_12_agent_to_agent_a2a_main_google_adk.py - A2A server with AgentCard, OAuth, and Starlette application

16. Resource-Aware Processing

• 16_11_resource_aware_google_adk.py - Google ADK QueryRouterAgent routing queries by complexity (Flash vs Pro)
• 16_12_resource_aware_openai.py - OpenAI prompt classifier routing to simple, reasoning, or search paths
• 16_13_resource_aware_openrouter.py - OpenRouter API with model selection and fallback strategies

17. Reasoning Techniques

• 17_11_reasoning_techniques_chain_of_thought.py - Chain of Thought prompting for step-by-step reasoning
• 17_12_reasoning_techniques_self_correction.py - Self-correction pattern for content refinement
• 17_21_reasoning_techniques_PALM_code_executors_google_adk.py - Google ADK with search and code execution agents

18. Guardrails

• 18_11_guardrails_crewai.py - CrewAI content policy enforcer with Pydantic validation for safety guardrails
• 18_12_guardrails_vertexai.py - Google ADK before_tool_callback for tool argument validation
• 18_21_guardrails_safety_agent_prompt.py - Safety guardrail prompt for filtering jailbreaking, harmful, and off-topic inputs

19. Evaluation and Monitoring

• 19_11_evaluation_and_monitoring_accuracy.py - Simple accuracy score calculation for agent responses
• 19_12_evaluation_and_monitoring_token_usage.py - LLMInteractionMonitor for tracking input/output token usage
• 19_13_evaluation_and_monitoring_token_helpfulness_llm.py - LLM-as-a-Judge rubric for evaluating survey question quality

Prerequisites

• Python 3.10+
• API keys for the respective LLM providers (OpenAI, Google Gemini, etc.)

Installation

pip install -r requirements.txt

Configuration

Create a .env file with your API keys:

OPENAI_API_KEY=your_openai_key
GOOGLE_API_KEY=your_google_key
DATABASE_ID=your_datastore_id  # Optional, for 5_5 examples

Security Note: Some scripts contain placeholder values like "YOUR_API_KEY". Replace these with your actual API keys or use environment variables before running.

Usage

Run individual scripts based on the pattern and framework you want to explore:

python 1_prompt_chaining_langchain.py
python 2_2_routing_google_adk.py
# etc.

License

MIT License - See LICENSE file for more details.