- install pycharm community edition ide (if you don't have one)
- install anaconda (if you don't want to use venv)
Join my Discord community for any questions!
Use Python 3.10, 3.11 or 3.12. The course was recorded on crewai==0.35.8 (July 2024), and requirements.txt pins numpy==1.26.4, which publishes wheels only up to Python 3.12.
To install for the beginning of the tutorial, type this in your terminal:
pip install crewai==0.35.8
The package is crewai. pip install crew installs an unrelated AMQP library. Keep the ==0.35.8 pin: current CrewAI (1.x) is a different API, and for example rejects the verbose=2 passed to Crew in main.py (it now takes True/False).
pip install -r requirements.txt and pip install 'crewai[tools]==0.35.8' no longer install. Both need lancedb 0.5.x (pinned directly in requirements.txt, and required as lancedb>=0.5.4,<0.6 by every crewai-tools 0.4.x release), and those lancedb releases are no longer on PyPI; the oldest one left is 0.14.0. That affects rag_pdf/, which imports crewai_tools (it also imports decouple, from pip install python-decouple, which is not in requirements.txt). main.py and tools-example.py only need crewai==0.35.8 (its dependencies include langchain-community and python-dotenv).
For current CrewAI code (CrewAI 1.15), see tylerprogramming/ai.
gpt-4 and gpt-3.5-turbo were shut down on 2026-10-23 (OpenAI deprecations page), so these files fail with model_not_found until you change the model name:
| File | Model in the code |
|---|---|
main.py |
OPENAI_MODEL_NAME = "gpt-4" |
.env.example (used by tools-example.py) |
OPENAI_MODEL_NAME=gpt-4 |
rag_pdf/agents.py |
ChatOpenAI(model_name="gpt-4", ...) and ChatOpenAI(model_name="gpt-3.5-turbo", ...) |
main - ollama - localai.py runs phi3:3.8b on a local Ollama server and is not affected.
Use gpt-5.4-mini as the replacement in all three places. CrewAI 0.35.8 calls OpenAI through langchain-openai 0.1.7, which always sends temperature (0.7 unless you set it). gpt-5.4-mini defaults to reasoning effort none, so it accepts that. OpenAI's listed substitutes, gpt-5.6-sol (for gpt-4) and gpt-5.6-terra (for gpt-3.5-turbo), default to reasoning effort medium, and OpenAI's docs say to remove temperature whenever reasoning effort is not none. Those need code changes this old LangChain version does not make easy.
Price per 1M input/output tokens: gpt-5.4-mini $0.75 / $4.50, gpt-5.6-terra $2 / $12, gpt-5.6-sol $4 / $20.