CrewAI is a leading agent framework in the 2026 AI agent engineering landscape (54.8k★).
When to use this skill
- Role-based multi-agent crews: Agent (role/goal/backstory) + Task + Crew + Process
- Flows: event-driven orchestration with @start/@listen/@router decorators for production pipelines
- YAML-first configuration (agents.yaml / tasks.yaml) scaffolded by the crewai CLI
- Memory and knowledge stores per agent/crew
- crewai-tools package + MCP tool support
- Official CrewAI Skills for coding agents (npx skills add crewaiinc/skills)
- Standalone — no LangChain dependency
- Ecosystem overlap: langgraph-workflow covers graph orchestration; crewai covers role/crew paradigm — complementary; langchain-bmad routes between framework skills
Installation
uv pip install crewai # extras: 'crewai[tools]'
Core API
crewai create crew my_crew
cd my_crew && crewai install
# edit src/my_crew/config/agents.yaml + tasks.yaml
crewai run
Instructions
Step 1: Initialize the environment
Verify you have Python installed and set up properly. Use the correct package manager (pip, npm, pnpm, uv) as specified in the installation section.
Step 2: Establish the agent flow
Architecture pattern: Two-layer: Crews (autonomous role-based collaboration, sequential/hierarchical process) and Flows (deterministic event-driven control flow that can embed crews)
Step 3: Implement tool integrations and policies
Integrate necessary tools, environment keys, and safety gates according to your architectural needs. Position in the ecosystem: Most-adopted role-based multi-agent framework; alternatives: agno (perf/platform focus), langgraph (graph control), autogen (maintenance mode)
Examples
Example 1: Minimal Agent Run
crewai create crew my_crew
cd my_crew && crewai install
# edit src/my_crew/config/agents.yaml + tasks.yaml
crewai run
Best practices
- Follow the primary installation commands for CrewAI to ensure dependencies are resolved correctly.
- Configure model providers and API keys securely before starting the runtime.
- Use appropriate sandboxing or permission isolation if executing AI-generated code.