Deep Agents Lifecycle

An interactive tutorial platform that teaches you how to build production-grade AI agents from scratch. 15 progressive chapters, each adding exactly one new concept — from "hello world" to a full multi-agent pipeline that generates charts, reports, and presentations.

What's Inside

deep-agents/
  deepagents-examples/     # 15 Python scripts — the actual tutorial
  backend/                  # FastAPI server that runs examples & streams output
  frontend/                 # Next.js interactive UI with live code viewer
  output/                   # Runtime output organized by chapter (ch01-ch15)
  workspace/                # Agent workspace (files the agent reads/writes)
  skills/                   # Skill files loaded by chapter 12+

The 15 Chapters

#	File	New Concept	What the Agent Does
01	01_hello_analyst.py	create_deep_agent()	Answers "what is a pivot table?"
02	02_pick_your_brain.py	Model swap	Same question on GPT vs Claude vs Gemini
03	03_give_it_eyes.py	Custom @tool	Reads CSV data with a custom tool
04	04_the_analyst.py	System prompt	Structured analysis with methodology
05	05_think_first.py	Planning (write_todos)	Plans before analyzing
06	06_files_on_disk.py	FilesystemBackend	Reads/writes real files on disk
07	07_safe_code_lab.py	Daytona sandbox	Runs pandas + matplotlib in isolation
08	08_analyst_team.py	Sub-agents	Data Wrangler + Chart Designer + Report Writer
09	09_it_remembers.py	Memory (checkpointer)	Remembers preferences across sessions
10	10_trust_gate.py	Human-in-the-loop	Pauses for approval on destructive ops
11	11_production_armor.py	Middleware	Retry, rate-limit, token tracking
12	12_domain_brain.py	Skills	Loads statistics playbook + chart design guide
13	13_typed_results.py	Structured output	Returns Pydantic AnalysisReport object
14	14_learning_agent.py	Persistent learning	Agent reads/writes AGENTS.md to learn from corrections
15	15_full_pipeline.py	Everything combined	CSV -> Charts -> PPT -> Word -> Excel

Quick Start

Prerequisites

• Python 3.11+
• Node.js 18+
• API keys for OpenAI (required), Anthropic (ch02), Daytona (ch07+)

1. Install Python dependencies

pip install deepagents langchain-daytona python-dotenv

2. Set up environment

cp .env.example .env
# Add your API keys to .env

Required keys:

OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...     # for chapter 02
DAYTONA_API_KEY=...               # for chapters 07+

3. Run individual chapters

cd deepagents-examples
python 01_hello_analyst.py
python 02_pick_your_brain.py
# ... work through all 15

4. Run the interactive UI

# Terminal 1 — Backend
cd backend
pip install -r requirements.txt
python server.py

# Terminal 2 — Frontend
cd frontend
npm install
npm run dev

Open http://localhost:3000 to access the interactive tutorial platform.

Interactive UI Features

• Learn Tab — Read through each chapter's concepts and code
• Lab Tab — Run any chapter's code live, watch the agent work in real-time with token stats, tool call tracking, and execution timing
• Explore Tab — Browse the codebase
• Agent Failure Arcade — 6 interactive games showing why standard agents fail (context rot, no planning, unsafe execution, amnesia, wrong tools, hallucination)

Output Structure

Each chapter writes its output to output/chXX/:

output/
  ch01/    # Simple text responses
  ch06/    # Files written to disk
  ch07/    # Charts generated in sandbox
  ch08/    # Multi-agent coordinated output
  ch15/    # Full pipeline: charts, Excel, PowerPoint, Word, Markdown

Tech Stack

• Agent Framework: deepagents (built on LangGraph)
• LLM Providers: OpenAI, Anthropic, Google (configurable per chapter)
• Sandbox: Daytona (isolated code execution)
• Backend: FastAPI + SSE streaming
• Frontend: Next.js 15, Tailwind CSS, Framer Motion

License

MIT