Introduction: Multi-Agent Systems Are the Next Frontier

LLMs are evolving from single-task bots to collaborative, multi-role agents.

With CrewAI, you can:

• Create specialized agents (planner, researcher, coder)

• Run them in workflows using Qwen3 models

• Add tools like browsers, APIs, vector DBs

• Host everything locally or in the cloud

In this guide, we’ll show you how to build a full-stack multi-agent system using Qwen3 + CrewAI.

---

1. What You Need

• ✅ Python 3.10+

• ✅ Qwen3 model (e.g., Qwen/Qwen1.5-14B-Chat)

• ✅ CrewAI library

• ✅ Optional tools: SerpAPI, LangChain, browser agent, Redis memory

Install dependencies:

bash
pip install crewai langchain openai transformers accelerate

---

2. Project Structure

arduino
qwen3-agents/
├── main.py
├── agents/
│   ├── planner.py
│   ├── researcher.py
│   └── coder.py
├── tools/
│   └── websearch.py
└── config.py

---

3. Create Your Agents

Example: agents/planner.py

python
from crewai import Agent

def PlannerAgent(llm):
    return Agent(
        role="Project Planner",
        goal="Create step-by-step instructions for any task",
        backstory="An expert at breaking down goals into logical sequences.",
        verbose=True,
        llm=llm
    )

Repeat for ResearcherAgent and CoderAgent.

---

4. Add Tools (Optional)

Example: Web Search tool (using SerpAPI or LangChain tools):

python
from langchain.tools import DuckDuckGoSearchRun
search_tool = DuckDuckGoSearchRun()

You can then pass this tool to your agents during setup.

---

5. Use Qwen3 as the LLM

In config.py:

python
from transformers import AutoModelForCausalLM, AutoTokenizer
from langchain.llms import HuggingFacePipeline
import torch
from transformers import pipeline

def load_qwen3_model():
    model_id = "Qwen/Qwen1.5-14B-Chat"
    tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
    model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", trust_remote_code=True)
    
    pipe = pipeline("text-generation", model=model, tokenizer=tokenizer, max_new_tokens=512)
    return HuggingFacePipeline(pipeline=pipe)

---

6. Create the Crew

python
from crewai import Crew
from agents.planner import PlannerAgent
from agents.researcher import ResearcherAgent
from agents.coder import CoderAgent
from config import load_qwen3_model

llm = load_qwen3_model()

crew = Crew(
    agents=[
        PlannerAgent(llm),
        ResearcherAgent(llm),
        CoderAgent(llm)
    ],
    verbose=True
)

result = crew.run("Build a Python app that scrapes weather data and displays it in a dashboard.")
print(result)

---

7. Add Memory or Persistence (Optional)

Use Redis, ChromaDB, or LangChain’s memory modules to give agents:

• Dialogue history

• Task memory

• Cross-session persistence

---

8. What You Can Build

---

Why Qwen3 Works So Well in CrewAI

• ✅ Large context window (up to 128K)

• ✅ Works with structured prompts (system + user roles)

• ✅ Instruction-tuned for role-based behavior

• ✅ Fast local inference with 7B/14B options

• ✅ Fully open-source (Apache 2.0)

---

Conclusion: Bring Autonomous Agents to Life with Qwen3

With Qwen3 + CrewAI, you get:

• Modular agents with custom goals

• Tool integration

• Agent collaboration

• Full control via local models

Now you can go beyond chatbots—and build fully open, intelligent systems that reason, plan, and build together.

---

Resources

• CrewAI GitHub

• LangChain Agents

• Qwen Models on Hugging Face

• Prompt Engineering with Qwen3