Official Resources

Tools Extension Example Workflows Community GUI Crews Concept Flows Concept Telemetry Installation Guide

Key Features

  • Crews - Role-Based Teams: Role-based teams of autonomous agents. Each agent has a role, goal, backstory, and optional tools. Supports sequential, hierarchical, parallel, consensual processes.
  • Flows - Event-Driven Pipelines: Event-driven, single-LLM pipelines. Annotate Python methods with @start, @listen, @router to build deterministic graphs.
  • Hybrid Orchestration: Switch freely between autonomous exploration (Crews) and strict deterministic pipelines (Flows). Flows can embed Crews when autonomy is required.
  • Built-in Observability: Shared runtime with built-in OpenTelemetry tracing and memory management for both Crews and Flows.
  • Core Components: Agent (LLM-powered actor), Tool (callable modules), Task (unit of work), Process (collaboration pattern), Telemetry (metrics and traces).

Code Examples

Installation and Setup

bash 
# 1. Install
pip install 'crewai[tools]'

# 2. Scaffold project
crewai create crew market_intelligence
cd market_intelligence

# 3. Add keys to .env
echo "OPENAI_API_KEY=sk-..." >> .env
echo "SERPER_API_KEY=..." >> .env

# 4. Run
crewai run

Agent Configuration

yaml 
# Example agent configuration
researcher:
  role: >
    {topic} Senior Data Researcher
  goal: >
    Uncover cutting-edge developments in {topic}
  backstory: >
    You're a seasoned researcher with a knack for uncovering the latest
    developments in {topic}.
writer:
  role: "AI Report Writer"
  goal: "Create a clear report on the research findings."
  backstory: >
    You are meticulous and excel at turning complex information into
    concise summaries.

Use Cases

  • End-to-end research pipelines - Researcher → Analyst → Writer
  • Code generation workflows - Architect → Coder → QA agent
  • Content creation - Trip planners, stock analysis reports, meeting assistants
  • Enterprise automation - Multi-level approvals, audit-trail required processes leveraging the Enterprise control plane

Pros & Cons

Advantages

  • Pure Python, zero LangChain dependency - Core wheel < 1 MB, 5-6× faster execution
  • Role-based design - Lower hallucination, increased clarity
  • Hybrid orchestration - Autonomous exploration (Crews) + deterministic pipelines (Flows)
  • Built-in observability - OpenTelemetry traces, metrics, Grafana dashboards
  • Rapid development - Project scaffolding in seconds
  • Growing ecosystem - 50+ tools, MCP bridge, GUI, Enterprise control plane

Disadvantages

  • Python-only SDK - No JavaScript, Go, or Rust clients
  • Smaller tool marketplace - <200 tools vs LangChain's 1k+
  • Memory-based state - No persistent storage layer yet
  • Experimental streaming - Token-level streaming not production-ready
  • Single-language teams - Requires wrappers for non-Python services

Future Outlook & Integrations

  • Enterprise Suite [In Development]: Crew Control-Plane SaaS (free tier available), On-prem Helm chart for Kubernetes, SSO, RBAC, and audit logs, SOC-2 audit scheduled August 2025
  • Tooling & Marketplace [Active]: Built-in tools, MCP adapter, and Composio integration. Upcoming marketplace for community-contributed tools
  • GUI & Low-code [Available]: CrewAI-Studio (community, Streamlit-based, cross-platform), Enterprise web playground for non-technical users
  • Deployment Targets [Planned]: Local, Docker, Kubernetes Operator (GA). Serverless workers (AWS Lambda, Cloud Run) on the roadmap