TLDR
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.
Status 2026-09-30
Python, MIT licence, ~59k GitHub stars; PyPI latest 1.15.23 (2026-09-28). Built from scratch, independent of LangChain (docs say so explicitly). Two layers: Crews (autonomous role-based agents) and Flows (event-driven workflows with state). Commercial: CrewAI AMP suite (managed deployment, observability, governance) with a free control plane. Needs Python 3.10 to <3.14. The LiteLLM-under-the-hood claim from 2024 is not re-verified. Note: the repo fetch showed “v0.102.0” as latest release, which conflicts with PyPI 1.15.x; PyPI is used here.
Comparison with TypeScript framework: mastra-vs-crewai-comparison.
Commercial offering: http://crewaiinc.com
Differences Between autogen, chatdev, and crewai
- AutoGen is good for conversational agents but lacks process concepts.
- ChatDev introduces processes but is rigid and not customizable for production.
- CrewAI combines the flexibility of AutoGen with ChatDev’s process approach without rigidity
Summarized March 2024 (AutoGen is now maintenance-only, see microsoft-agent-framework) from Gao Dalie (高達烈) | Instagram, TikTok | Linktree on Medium
Tutorials, Blogs
- Brian Hancock: Crew AI advanced tutorial
- Gao Dalie: CrewAI shared memory
- Replit: CrewAI AI agents in the cloud with Replit Company Spotlight: CrewAI