Skip to content

Exercise 1: Same Agent, Two Frameworks (LangGraph + CrewAI)

Pairs with Stage 4 — Workflow Graphs & Agent Frameworks Exercise 1.

🎓 How to use this: First run the provided starter.py (python starter.py), then change exactly one small thing and run the existing test again: python test.py. If the test fails, undo or fix that one change and try again. You do not need to rename the file or rewrite the whole solution. See docs/HOW_TO_USE.md for the full method.

📚 Want the chapter-length version? The starter in this folder is an illustrative build focused on the core pattern plus two SDK paths — it is not in-depth teaching material. Recommended for depth: - datawhalechina/hello-agents ⭐ the most complete Chinese-language course out there — chapter by chapter, plus 16 production capabilities. This exercise maps to hello-agents' framework comparison / orchestration chapter - LangGraph Quickstart + CrewAI official docs - Full references in Stage 4 Curated Projects

Task

A minimal search + summarize agent:

  • Given a query (e.g. "summarize Taipei")
  • Agent uses a search tool to hit a knowledge base
  • LLM summarizes the result in 1-2 sentences

Built once in LangGraph and once in CrewAI — compare styles.

How to run — two paths + two frameworks

⚠️ Give each exercise its own Python 3.11 .venv. Do not install all five requirements.txt files together. They demonstrate different frameworks whose dependency ranges can conflict.

Path A (default, free, local)

py -3.11 -m venv .venv
.\.venv\Scripts\python.exe -m pip install -r requirements.txt
ollama pull qwen2.5:3b
ollama serve

.\.venv\Scripts\python.exe starter.py # LangGraph + Ollama
.\.venv\Scripts\python.exe starter_crewai.py # CrewAI + Ollama comparison

Budget: the model API costs $0. Your computer, memory, electricity, and download time are not free.

Path B (Anthropic, compare a cloud result)

$env:ANTHROPIC_API_KEY="sk-ant-..."
.\.venv\Scripts\python.exe starter_anthropic.py # LangGraph + Claude

Pinned default: claude-haiku-4-5-20251001. Haiku 4.5 costs $1 per million input tokens and $5 per million output tokens. A request with 2,000 input + 1,000 output tokens costs 2,000 / 1,000,000 × $1 + 1,000 / 1,000,000 × $5 = $0.007. A framework may make more than one request, so set a provider spend limit of $0.05 for this exercise. This is an estimate, not a billing promise.

macOS/Linux commands and verification information 
python3.11 -m venv .venv
./.venv/bin/python -m pip install -r requirements.txt
export ANTHROPIC_API_KEY="sk-ant-..."
./.venv/bin/python test.py

Official sources: LangGraph overview | CrewAI docs | Anthropic pricing

Packages, model IDs, prices, and official links verified: 2026-08-28 UTC.

Validate the logic (mock-based)

.\.venv\Scripts\python.exe test.py # LangGraph behavior
.\.venv\Scripts\python.exe test_anthropic.py # Anthropic-path behavior
.\.venv\Scripts\python.exe test_crewai.py # CrewAI behavior

Side-by-side framework comparison

Dimension LangGraph CrewAI
Core abstraction StateGraph + node + edge Agent + Task + Crew
Mental model "How does state flow?" "Who plays what role?"
Loop control Explicit conditional edges Hidden inside Crew.kickoff()
Debug path Inspect graph state and checkpoints Inspect task output and verbose logs
Useful when You need explicit state and branches You want to express role/task collaboration quickly
Learning curve Medium-high Low

LangGraph style (condensed)

g = StateGraph(State)
g.add_node("agent", agent_node)
g.add_node("tools", tool_node)
g.add_conditional_edges("agent", should_continue, {"tools": "tools", END: END})
g.add_edge("tools", "agent")

"I tell the system explicitly: state shape, nodes, edges, branching via should_continue."

CrewAI style (condensed)

researcher = Agent(role="Researcher", goal="...", tools=[search], llm=MODEL)
task = Task(description=query, expected_output="...", agent=researcher)
crew = Crew(agents=[researcher], tasks=[task])
crew.kickoff()

"I describe: who plays this role, what task, what tools. Framework decides how to run."

What to observe

  1. Abstraction cost: CrewAI hides more, writes less code; but stack depth grows when debugging
  2. Small-model behavior: test both paths; role descriptions, tool schemas, and task length can all change the result
  3. Controllability: LangGraph exposes state transitions; CrewAI is "result-oriented"
  4. When to pick: try LangGraph when you need to inspect each state transition; try CrewAI when roles are the clearest first description, then measure on your own task

Common pitfalls

  • LangGraph bind_tools: must llm.bind_tools([search]) to expose tool schema. Without it the model doesn't know the tool exists
  • CrewAI model spec: use LiteLLM format ("ollama/qwen2.5:3b", not "qwen2.5:3b"). A wrong provider prefix can select a different backend, so print and verify the setting before a run
  • CrewAI return type: crew.kickoff() returns a CrewOutput object; str(result) to get text. Bare print(result) may show repr

Want smarter answers?

$env:MODEL="claude-sonnet-5"; .\.venv\Scripts\python.exe starter_anthropic.py
$env:MODEL="qwen2.5:7b"; .\.venv\Scripts\python.exe starter.py

Extensions

  • Streaming: use LangGraph graph.stream(...); for CrewAI, construct Crew(..., stream=True) and then call crew.kickoff()
  • Checkpointing: LangGraph + MemorySaver for time-travel debug
  • Human-in-the-loop: see Exercise 3