CrewAI
CrewAI is an open-source Python framework for building multi-agent workflows that go from prototype to production. It combines two ideas: Crews, teams of role-based agents with goals, tools and tasks that collaborate and delegate autonomously, and Flows, event-driven workflows with state, branching and routing that give you precise control. You can use either on its own or combine them, keeping application logic in regular Python.
It suits developers who need more than a single prompt or chatbot: multi-step work with specialised agents, tool use, structured outputs, human review steps, or workflows that mix autonomous reasoning with explicit business logic. Examples in the repository include writing job descriptions, trip planning and stock analysis. Agents can use tools, memory, knowledge sources, checkpointing, async execution and MCP or A2A integrations, and local models are supported through tools such as Ollama and LM Studio.
Getting started: CrewAI requires Python 3.10 to 3.13 and uses uv for dependency management. Install uv, run `uv tool install crewai`, then `crewai create crew` to scaffold a project and `crewai run` to execute it. CrewAI is standalone and does not depend on LangChain.
Features
Why choose it
Trade-offs
Maintenance check
- Last commit
- 2026-10-23
- Commits, last 90 days
- 100+
- Releases, last 90 days
- 24
- Latest release
- 1.15.29 (2026-10-23)
- Open issues
- 824
- Recent issues with at least one comment
- 7/18
Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.
Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.
Most discussed open issues
- #5802 Tool re-execution on task retry has no idempotency guard — duplicate payments, emails, trades possible212 comments · 2026-05-14
- #5888 [FEATURE]:Governance middleware hook for tool call authorization148 comments · 2026-05-21
- #6025 [FEATURE] Runtime release-control mediation layer before agent/tool execution127 comments · 2026-06-03
- #6414 [FEATURE] Native deterministic guardrail to prevent infinite agent delegation and tool loops51 comments · 2026-07-01
- #5556 [FEATURE] Add pre-execution validation for agent-to-agent actions34 comments · 2026-04-20
Issue titles are quoted from GitHub. Open the link to check whether a problem still applies to the version you use.
Who it fits
- You want a Python framework that combines autonomous role-based agent teams (Crews) with deterministic event-driven workflows (Flows).
- You need built-in checkpointing, structured outputs, human-in-the-loop input, and MCP/A2A support under an MIT license.
- Your stack is .NET or multi-language; Microsoft Agent Framework or Google ADK cover non-Python runtimes.
- You want a lightweight, single-agent loop focused purely on strict Pydantic type validation; Pydantic AI is simpler for that.
How it differs
vs Microsoft AutoGen★ 61.6kCrewAI is actively developed around role-based Crews and event-driven Flows in Python (MIT). Microsoft AutoGen is now in maintenance mode, with Microsoft directing new projects to Microsoft Agent Framework.
vs Pydantic AI★ 20.9kCrewAI focuses on multi-agent role delegation and multi-step Flows. Pydantic AI focuses on end-to-end type safety, dependency injection, and validated structured outputs in a Python agent loop.
Compatibility
Quick start
Use cases
Alternatives
Featured in these lists
Comments
No comments yet. Be the first!