1Foundation
Creating Talking Agents
Basic talking agents and orchestration.
SETUP
Getting a Gemini API Key
1
Sign in to AI Studio
Go to Google AI Studio and log in with your Google account.
2
Open API Keys
Select API Keys from the left navigation panel.
3
Create an API Key
Click Get API Key and connect a Google Cloud project.
4
Store Securely
Save the key in environment variables or configuration files.
INDEX
Repository Index
Each script introduces one new mental model. Read them in sequence to understand how modern agentic systems evolve from simple prompts into orchestration architectures.
2Memory
Agents with Personality and Memory
Identity, memory, and fact extraction.
3Routing
Which Agent Should Speak?
Moderator controls conversation routing.
4Tools
Agents Using Tools
Tool-calling agents using external functions.
5JSON
Agents Who Return JSON
Structured outputs and machine-readable reasoning.
6Handoff
Reasoning Agents and Handoff
Agents reason and route control dynamically.
7Supervisor
Supervisor Pattern
Supervisor routes work to specialist agents.
8Parallel
Parallel Agents
Fan-out and fan-in reasoning systems.
9Debate
Debate Pattern
Competing agents and judge evaluation.
10Persistent Memory
Long-Term Memory
Persistent database-backed memory systems.
11Shared Memory
Agents with Shared Memory
Collaborative multi-agent memory architecture.
12Extraction
Knowledge Extraction
Transform text into structured entities.
13Embeddings
Chunking and Embedding
Prepare documents for semantic retrieval.
14Vector Search
Vector Search Before Generation
Retrieve vectors before answering.
15RAG
Basic RAG Agent
Retrieval-augmented generation basics.
16Agentic RAG
Agentic RAG
Agent decides when retrieval is needed.
17Multi-hop
Multi-hop Retrieval
Retrieval across multiple reasoning steps.
18MCP
MCP Protocol Walkthrough
Understanding the Model Context Protocol.
19MCP Client
Connecting to an MCP Server
Client-side MCP communication.
20MCP Server
Creating Your Own Tool Server
Build custom MCP-powered servers.
21Integration
Agent + MCP Full Integration
End-to-end architecture with MCP tools.
GUIDE
How to Read the Scripts
Find the Agent Class
Start with the class that wraps Gemini. That is the mind of the pattern.
Trace the State
Look for memory, facts, retrieved context, JSON outputs, and tool results.
Read the Orchestrator
The orchestrator tells you how control moves between agents.
Inspect Prompts
System prompts define roles, routing, constraints, and structured behavior.
Run and Modify
Change one variable at a time: role, memory, tool, retrieval, or routing.
MENTAL MODELS
Pattern Views
Core loop
Agent = prompt + memory + model call
The earliest scripts teach the basic loop: receive input, build context, call Gemini, store response.
Control
Orchestration decides movement
Alternation, moderation, supervisor routing, and handoff are all control-flow patterns.
Grounding
Retrieval gives the agent evidence
Chunking, embedding, vector search, RAG, and multi-hop retrieval prevent blind guessing.
Capability
MCP turns tools into an ecosystem
The MCP patterns separate reasoning from capability, letting agents call external tool servers.
Final Thoughts
This cookbook moves from simple conversations to serious agent architecture: memory, tools, supervision, retrieval, MCP, and full integration.
Read the scripts in order. Each one adds a new capability to the previous mental model. By the end, you are no longer just prompting a model — you are designing an intelligent system.