CrewAI is great! ... but

... getting started is hard. I'm a seasoned developer and using AzureOpenAI with CrewAI when it's always being updated is tricky. This repo is designed to demonstrate how to use the (almost) latest version of CrewAI with AzureOpenAI.

Explanation

• This devcontainer has been carefully set up to use the right python and sqlite3 versions.
• This project has been set up as per the CrewAI Quickstart.
• Effectively, only the azure_llm.py file has been added
• crew.py has been updated to reference the azure_llm.py file.

Instructions

1. Clone this repo
2. Run in GitHub codespaces (or locally, but instructions are for codespaces)
3. In /latest_ai_development/src/latest_ai_development/azure_llm.py replace the AZURE_API_KEY and AZURE_API_BASE and AZURE_API_VERSION and DEPLOYMENT_NAME with your own values.
4. As per the CrewAI Quickstart, cd into the latest_ai_development directory and run crewai install
5. Run crewai run to start the CrewAI crew

Notes

I've set up GitHub codespaces as a few things: Chroma requires a modern version of sqlite3. At time of writing (Oct 2024) the universal GitHub codespace comes with Ubuntu LTS 20 and very old version of sqlite3 and uses conda for python. Very confusing; even if we install python again, it will still take the old sqlite3 version. Also, if we swap the universal codespaces image with the python docker image, it also has an old version of sqlite3. So, we need to use a modern Ubuntu version as our base image and install python3 in the codespace using features.

I've added a requirements.txt which is the most important file as it tracks the latest working version of CrewAI. It will lag behind the actual latest version of CrewAI until I've had time to test the very latest version actually works.