CrewAI is a standalone Python framework for orchestrating multi-agent workflows, built around two complementary primitives. "Crews" model teams of role-based agents — each with a role, goal, backstory, and tools — that collaborate and delegate autonomously, using either a sequential process or a hierarchical one where a manager agent coordinates and validates the work. "Flows" are event-driven workflows that give explicit, deterministic control over execution: structured state, conditional branching, routing, and the ability to call Crews as steps.
The framework is designed to combine the two so autonomous agent collaboration can be wrapped in controlled, production-oriented pipelines. Agents and tasks are typically defined in YAML config plus Python, and the framework supports tools, memory, knowledge, checkpointing, async execution, structured outputs, human-in-the-loop review, and MCP/A2A. It is LLM-agnostic (OpenAI by default, with support for other providers and local models via Ollama/LM Studio), and collects anonymous usage telemetry by default that can be disabled. A separate commercial "AMP Suite" adds a managed control plane with tracing, observability, governance, and enterprise deployment.