CrewAI

CrewAI is a lean, standalone Python framework for building and orchestrating multi-agent AI systems, released under the MIT License. It is built entirely from scratch—independent of LangChain or other agent frameworks—and pairs an open-source core with an enterprise Agent Management Platform (AMP) for production deployments. The framework is actively maintained, with the latest release (v1.14.5) published in May 2026.

What It Is

CrewAI provides two complementary programming models for multi-agent automation: Crews and Flows. Crews are teams of autonomous, role-playing AI agents that collaborate dynamically to accomplish complex tasks through delegation and shared goals. Flows are event-driven, production-ready workflows that give developers fine-grained, deterministic control over execution paths, state management, and conditional branching. The two models are designed to be combined, letting builders balance autonomy with precision in the same application.

The Two-Layer Architecture: OSS and AMP

The product ships in two distinct layers:

• CrewAI OSS: The open-source Python framework installable via uv pip install crewai. It exposes high-level abstractions (agents, tasks, crews, flows) and low-level APIs for deep customization, supports sequential and hierarchical processes, and connects to any LLM including local models via Ollama.
• CrewAI AMP (Agent Management Platform): A commercial suite built on top of the OSS framework. AMP includes CrewAI Studio (a visual no-code editor with an AI copilot), real-time workflow tracing, automated and human-in-the-loop agent training, role-based access control, serverless containers, and centralized monitoring. AMP is available as a cloud-hosted service (AMP Cloud) or as a self-hosted deployment on private infrastructure including on-premises, AWS, Azure, or GCP (AMP Factory).

Developer Workflow and Setup

Getting started with the OSS framework requires Python 3.10–3.13 and uses the uv package manager. The CLI scaffolds a full project structure:

crewai create crew <project_name>

This generates agents.yaml, tasks.yaml, crew.py, and main.py files. Agents are defined with a role, goal, and backstory; tasks specify descriptions, expected outputs, and assigned agents. Crews can run sequentially or hierarchically, with a manager agent automatically coordinating delegation in hierarchical mode. Flows use Python decorators (@start, @listen, @router) to wire event-driven pipelines and support logical operators (or_, and_) for complex branching.

LLM and Tool Integrations

CrewAI connects to a wide range of language models through its LLM Connections layer, defaulting to OpenAI but supporting Anthropic, Google, local models via Ollama and LM Studio, and IBM WatsonX. The AMP Studio ships with pre-built tool integrations including Gmail, Microsoft Teams, Notion, HubSpot, Salesforce, and Slack. Custom tools can be added via the API layer. The framework also publishes an MCP server for live documentation queries and official "skills" packages for AI coding agents (Claude Code, Cursor, Windsurf, Codex).

Update: Version 1.14.5

The latest release is v1.14.5, published May 18, 2026, on a rapid release cadence. The GitHub repository shows 51,000+ stars and 7,100+ forks as of mid-2026. The README states that over 100,000 developers have been certified through community courses at learn.crewai.com (a vendor-published figure). The homepage states the platform runs 450,000,000+ agentic workflows per month and sees 4,000+ sign-ups per week—figures published by CrewAI, Inc. The product direction is clearly toward enterprise adoption, with AMP adding governance, observability, and deployment flexibility on top of the open-source core.