Open-source project for orchestrating autonomous AI agents.
Similar to Microsoft’s AutoGen but easier to use and production-oriented.
Trending on Product Hunt with over 4,000 GitHub stars.
How Crew AI Works
Role-based agent design with each agent as an LLM (Large Language Model) with specific tools and tasks.
Agents collaborate to complete tasks as a team.
Compatible with both open-source and proprietary LLMs.
Using Crew AI in Google Colab
Tutorial on setting up Crew AI in a Google Colab notebook for free.
Installation of necessary packages including Crew AI, DuckDuckGo search, and Langchain Google generative AI.
Importing OS package and setting up Gemini Pro LLM for the example.
Setting Up Agents and Tasks
Define agent roles, goals, and backstories.
Assign LLMs and tools to agents.
Create tasks for agents to perform.
Form a crew with agents and assign tasks for sequential execution.
Execution and Results
Crew AI uses Langchain for internal processing.
Agents consider whether to use tools and execute actions accordingly.
Example output includes a report and a blog post draft, with some inaccuracies noted.
Conclusion
Crew AI is under active development with potential for future enhancements.
Encourages feedback on autonomous AI agent frameworks.
Example Code Usage in Google Colab
# Install packages !pip install crew-ai duckduckgo-search lang-chain-google-generative-ai # Import packages import os from crew_ai import Agent, Task, Crew, Process # Define agents with roles, goals, and backstories researcher = Agent(role='Senior Research Analyst', goal='Uncover AI developments', backstory='...', tools=[search_tool], llm='Gemini Pro') writer = Agent(role='Writer', goal='Develop engaging blog post', backstory='...', tools=[], llm='Gemini Pro') # Define tasks task_analysis = Task(description='Analyze AI advancements in 2024', agent_role='Researcher') task_blog_post = Task(description='Write blog post on AI advancements', agent_role='Writer') # Create crew and assign tasks crew = Crew(agents=[researcher, writer], tasks=[task_analysis, task_blog_post]) # Execute tasks crew.execute()
Note: The above code is a simplified representation based on the provided text and does not include all details such as API keys or specific package functions.