write-a-book-with-flows → Lyzr ADK

Translation: CrewAI Flows → Lyzr ADK
Source: https://github.com/crewAIInc/crewAI-examples/tree/main/flows/write_a_book_with_flows
Generated by: Framework Translator Agent v2.0.0

---

What this does

An AI book-writer pipeline that:

1. Researches a topic and generates a structured chapter outline
2. Writes all chapters in parallel (~3 000 words each, Markdown)
3. Joins and saves the complete book to a .md file on disk

The original CrewAI code used two Crews orchestrated by a Flow; this translation uses four Lyzr ADK agents orchestrated by a plain Python async function that preserves identical parallel chapter-writing behaviour.

---

Agent Map

Lyzr Agent	Source CrewAI Agent	Crew
outline_researcher	researcher	OutlineCrew
book_outliner	outliner	OutlineCrew
chapter_researcher	researcher	WriteBookChapterCrew
chapter_writer	writer	WriteBookChapterCrew

---

Setup

1. Install dependencies

pip install lyzr-adk requests pydantic

2. Set environment variables

export LYZR_API_KEY="sk-..."        # from https://studio.lyzr.ai → Settings → API Keys
export OPENAI_API_KEY="sk-..."      # required by openai/gpt-4o inside Lyzr
export SERPER_API_KEY="..."         # from https://serper.dev — used by web_search tool

3. Run

python3 write_a_book_lyzr_adk.py

The script will:

• Create 4 agents in your Lyzr Studio workspace (one-time)
• Run the book generation flow
• Save the finished book as The_Current_State_of_AI_in_July_2025.md in the current directory

---

Translation Notes

#	Source (CrewAI)	Target (Lyzr ADK)	Fidelity
1	Flow[BookState] with @start/@listen	Plain async Python orchestration function	Adapted
2	crewai_tools.SerperDevTool	web_search() function calling Serper.dev REST API	Adapted
3	output_pydantic=BookOutline / output_pydantic=Chapter (CrewAI task)	response_model=BookOutline / response_model=Chapter (Lyzr agent.run())	Direct
4	@persist durable Flow state	Not supported — crash = restart from Stage 1	Lossy
5	asyncio.gather() parallel chapters	asyncio.gather() preserved identically	Direct
6	Process.sequential inside each Crew	Sequential agent.run() calls within each stage	Direct
7	crew.kickoff(inputs={...})	agent.run(prompt)	Adapted
8	YAML agents.yaml / tasks.yaml configs	Inline studio.create_agent(role=, goal=, instructions=)	Adapted

---

Architecture

[main()]
   │
   ▼
[create_all_agents()]   ← POST /v3/agents/ × 4  (runs once at startup)
   │
   ▼
[generate_book_outline()]          ← Stage 1 (sequential)
   │  outline_researcher.run()  →  research findings
   │  book_outliner.run()       →  BookOutline (JSON → Pydantic)
   │
   ▼
[write_chapters()]                 ← Stage 2 (parallel via asyncio.gather)
   │  for each ChapterOutline:
   │    chapter_researcher.run()  →  research findings
   │    chapter_writer.run()      →  Chapter (Markdown)
   │
   ▼
[join_and_save_chapter()]          ← Stage 3 (sequential)
      Combine all chapters → write ./{Book_Title}.md

---

Files

File	Description
write_a_book_lyzr_adk.py	Main translated script — run this
README.md	This file