This repository is an example of a local setup for using agentic workflow using CrewAI and Ollama models. You can change the goal, role, and backstory of the agents, as well as the tasks' descriptions based on your desired output.
This project uses a free API formality key from OpenAI, but you can replace it with your paid API key from other providers.
- Set up a virtual environment using the
pyvenv.cfgsetting. - Configure your environment using the
dotenvor.envfile in your bin directory with Ollama endpoints and your choice of the large language model. - Create a Model File to specify the base LLM and set the PARAMETERs and the SYSTEM message to shape the behaviour of the LLM.
- Create a shell file to pull the information from the Model File and then execute it via your terminal.
- Make the necessary changes to the Python file based on your project goals, then run it.
This is a simple setup project that employs only two agents, each performing one task. You may set as many refiners as you need. You can also enhance the quality of the responses by setting allow delegation to True to allow agents to solve more complex problems using a chain of internal enquiries. This project uses a sequential task execution but you can set up a hierarchical one.
