CrewAI: Autonomous Agents and Workflows

Summary

CrewAI is a standalone Python framework for orchestrating autonomous, role-based AI agents and event-driven workflows.
Provides two complementary models: Crews (agent teams with autonomy) and Flows (production-ready, event-driven workflows).
Emphasizes high performance, low resource use, deep customization, and enterprise readiness.
Offers an AOP Suite (Crew Control Plane) for observability, security, integrations, and 24/7 enterprise support.
Open-source (MIT) with telemetry (anonymous by default) and optional deeper sharing via share_crew.

Standalone & Lean: Built from scratch with no LangChain dependency.
Crews: Autonomous, role-based agent collaboration and dynamic task delegation.
Flows: Precise event-driven control, structured state, conditional routing, and production integration.
Integration: Supports many LLMs and local models (e.g., Ollama); seamless tool and API integration.
Customization: Fine-grained control of workflows, agent prompts, behaviors, and execution logic.
Enterprise Suite: Unified control plane, tracing/observability, security, on-premise or cloud deployment.
Community & Learning: Documentation, examples, and courses; over 100,000 certified developers.

Requirements: Python >=3.10 and <3.14.
Install: pip install crewai (optional tools: pip install 'crewai[tools]').
Project scaffold: use CLI command crewai create crew <project_name>.
Key project files: src/<project>/crew.py, main.py, config/agents.yaml, config/tasks.yaml, .env.
Run: set API keys in .env, then run crewai run or python src/<project>/main.py.
Troubleshooting: common tiktoken/compilation issues addressed with optional installs and build tool notes.

File/Item	Purpose
src/<project>/crew.py	Defines Crew, agents, tasks, and process (sequential or hierarchical)
src/<project>/main.py	Entry point; prepares inputs and kicks off the Crew
src/<project>/config/agents.yaml	Agent roles, goals, and backstories configuration
src/<project>/config/tasks.yaml	Task descriptions, expected outputs, agent assignment
.env	Environment variables (OpenAI, Serper.dev, etc.)

Agents defined via decorators (@agent) and tasks via @task produce automatic wiring in Crew class.
Two process modes: sequential (explicit task order) and hierarchical (manager coordinates delegation/validation).
Flows support decorators (@start, @listen, @router) and logical combinators or_/and_ for complex routing.
Example flow integrates structured state (Pydantic BaseModel) with conditional routing based on state values.

Tracing & Observability: real-time metrics, logs, traces.
Unified Management: central control for agents, workflows, and scaling.
Integrations: connect with enterprise systems, data sources, cloud infra.
Security & Compliance: built-in advanced security features.
Insights & Support: analytics, reporting, and 24/7 enterprise support.
Deployment Options: cloud and on-premise supported; partial free trial available.

Default: anonymous telemetry collects metadata (CrewAI version, Python version, OS, number of agents/tasks, process type, model used, features used).
Sensitive content (prompts, outputs, secrets) is not collected unless share_crew=True.
Users can disable telemetry by setting OTEL_SDK_DISABLED=true.
Optional opt-in: share_crew collects detailed crew/task attributes for product improvement.

Example repos: CrewAI-examples (landing page generator, trip planner, stock analysis, job postings).
Tutorials: Getting Started and Quick Tutorial videos linked in docs.
Courses: deeplearning.ai short courses on multi-agent systems with CrewAI.
Docs: official docs and LLM connection guides explain integrations and local model setups.

Framework design: intentionally independent from LangChain and similar frameworks.
License: MIT, open-source and community-driven.
Telemetry: anonymous by default; detailed sharing requires explicit opt-in.