{"id":"crewai","name":"crewai-multi-agent","summary":"自律的なAI協働のためのマルチエージェントオーケストレーションフレームワーク。複雑なタスクで協力する専門的なエージェントのチーム構築、メモリを使ったロールベースのエージェント協働が必要な場合、または順次・階層的な実行を必要とする本番ワークフローの際に利用します。","body":"# CrewAI - Multi-Agent Orchestration Framework\n\nBuild teams of autonomous AI agents that collaborate to solve complex tasks.\n\n## When to use CrewAI\n\n**Use CrewAI when:**\n- Building multi-agent systems with specialized roles\n- Need autonomous collaboration between agents\n- Want role-based task delegation (researcher, writer, analyst)\n- Require sequential or hierarchical process execution\n- Building production workflows with memory and observability\n- Need simpler setup than LangChain/LangGraph\n\n**Key features:**\n- **Standalone**: No LangChain dependencies, lean footprint\n- **Role-based**: Agents have roles, goals, and backstories\n- **Dual paradigm**: Crews (autonomous) + Flows (event-driven)\n- **50+ tools**: Web scraping, search, databases, AI services\n- **Memory**: Short-term, long-term, and entity memory\n- **Production-ready**: Tracing, enterprise features\n\n**Use alternatives instead:**\n- **LangChain**: General-purpose LLM apps, RAG pipelines\n- **LangGraph**: Complex stateful workflows with cycles\n- **AutoGen**: Microsoft ecosystem, multi-agent conversations\n- **LlamaIndex**: Document Q&A, knowledge retrieval\n\n## Quick start\n\n### Installation\n\n```bash\n# Core framework\npip install crewai\n\n# With 50+ built-in tools\npip install 'crewai[tools]'\n```\n\n### Create project with CLI\n\n```bash\n# Create new crew project\ncrewai create crew my_project\ncd my_project\n\n# Install dependencies\ncrewai install\n\n# Run the crew\ncrewai run\n```\n\n### Simple crew (code-only)\n\n```python\nfrom crewai import Agent, Task, Crew, Process\n\n# 1. Define agents\nresearcher = Agent(\n    role=\"Senior Research Analyst\",\n    goal=\"Discover cutting-edge developments in AI\",\n    backstory=\"You are an expert analyst with a keen eye for emerging trends.\",\n    verbose=True\n)\n\nwriter = Agent(\n    role=\"Technical Writer\",\n    goal=\"Create clear, engaging content about technical topics\",\n    backstory=\"You excel at explaining complex concepts to general audiences.\",\n    verbose=True\n)\n\n# 2. Define tasks\nresearch_task = Task(\n    description=\"Research the latest developments in {topic}. Find 5 key trends.\",\n    expected_output=\"A detailed report with 5 bullet points on key trends.\",\n    agent=researcher\n)\n\nwrite_task = Task(\n    description=\"Write a blog post based on the research findings.\",\n    expected_output=\"A 500-word blog post in markdown format.\",\n    agent=writer,\n    context=[research_task]  # Uses research output\n)\n\n# 3. Create and run crew\ncrew = Crew(\n    agents=[researcher, writer],\n    tasks=[research_task, write_task],\n    process=Process.sequential,  # Tasks run in order\n    verbose=True\n)\n\n# 4. Execute\nresult = crew.kickoff(inputs={\"topic\": \"AI Agents\"})\nprint(result.raw)\n```\n\n## Core concepts\n\n### Agents - Autonomous workers\n\n```python\nfrom crewai import Agent\n\nagent = Agent(\n    role=\"Data Scientist\",                    # Job title/role\n    goal=\"Analyze data to find insights\",     # What they aim to achieve\n    backstory=\"PhD in statistics...\",         # Background context\n    llm=\"gpt-4o\",                             # LLM to use\n    tools=[],                                 # Tools available\n    memory=True,                              # Enable memory\n    verbose=True,                             # Show reasoning\n    allow_delegation=True,                    # Can delegate to others\n    max_iter=15,                              # Max reasoning iterations\n    max_rpm=10                                # Rate limit\n)\n```\n\n### Tasks - Units of work\n\n```python\nfrom crewai import Task\n\ntask = Task(\n    description=\"Analyze the sales data for Q4 2024. {context}\",\n    expected_output=\"A summary report with key metrics and trends.\",\n    agent=analyst,                            # Assigned agent\n    context=[previous_task],                  # Input from other tasks\n    output_file=\"report.md\",                  # Save to file\n    async_execution=False,                    # Run synchronously\n    human_input=False                         # No human approval needed\n)\n```\n\n### Crews - Teams of agents\n\n```python\nfrom crewai import Crew, Process\n\ncrew = Crew(\n    agents=[researcher, writer, editor],      # Team members\n    tasks=[research, write, edit],            # Tasks to complete\n    process=Process.sequential,               # Or Process.hierarchical\n    verbose=True,\n    memory=True,                              # Enable crew memory\n    cache=True,                               # Cache tool results\n    max_rpm=10,                               # Rate limit\n    share_crew=False                          # Opt-in telemetry\n)\n\n# Execute with inputs\nresult = crew.kickoff(inputs={\"topic\": \"AI trends\"})\n\n# Access results\nprint(result.raw)                             # Final output\nprint(result.tasks_output)                    # All task outputs\nprint(result.token_usage)                     # Token consumption\n```\n\n## Process types\n\n### Sequential (default)\n\nTasks execute in order, each agent completing their task before the next:\n\n```python\ncrew = Crew(\n    agents=[researcher, writer],\n    tasks=[research_task, write_task],\n    process=Process.sequential  # Task 1 → Task 2 → Task 3\n)\n```\n\n### Hierarchical\n\nAuto-creates a manager agent that delegates and coordinates:\n\n```python\ncrew = Crew(\n    agents=[researcher, writer, analyst],\n    tasks=[research_task, write_task, analyze_task],\n    process=Process.hierarchical,  # Manager delegates tasks\n    manager_llm=\"gpt-4o\"           # LLM for manager\n)\n```\n\n## Using tools\n\n### Built-in tools (50+)\n\n```bash\npip install 'crewai[tools]'\n```\n\n```python\nfrom crewai_tools import (\n    SerperDevTool,           # Web search\n    ScrapeWebsiteTool,       # Web scraping\n    FileReadTool,            # Read files\n    PDFSearchTool,           # Search PDFs\n    WebsiteSearchTool,       # Search websites\n    CodeDocsSearchTool,      # Search code docs\n    YoutubeVideoSearchTool,  # Search YouTube\n)\n\n# Assign tools to agent\nresearcher = Agent(\n    role=\"Researcher\",\n    goal=\"Find accurate information\",\n    backstory=\"Expert at finding data online.\",\n    tools=[SerperDevTool(), ScrapeWebsiteTool()]\n)\n```\n\n### Custom tools\n\n```python\nfrom crewai.tools import BaseTool\nfrom pydantic import Field\n\nclass CalculatorTool(BaseTool):\n    name: str = \"Calculator\"\n    description: str = \"Performs mathematical calculations. Input: expression\"\n\n    def _run(self, expression: str) -> str:\n        try:\n            result = eval(expression)\n            return f\"Result: {result}\"\n        except Exception as e:\n            return f\"Error: {str(e)}\"\n\n# Use custom tool\nagent = Agent(\n    role=\"Analyst\",\n    goal=\"Perform calculations\",\n    tools=[CalculatorTool()]\n)\n```\n\n## YAML configuration (recommended)\n\n### Project structure\n\n```\nmy_project/\n├── src/my_project/\n│   ├── config/\n│   │   ├── agents.yaml    # Agent definitions\n│   │   └── tasks.yaml     # Task definitions\n│   ├── crew.py            # Crew assembly\n│   └── main.py            # Entry point\n└── pyproject.toml\n```\n\n### agents.yaml\n\n```yaml\nresearcher:\n  role: \"{topic} Senior Data Researcher\"\n  goal: \"Uncover cutting-edge developments in {topic}\"\n  backstory: >\n    You're a seasoned researcher with a knack for uncovering\n    the latest developments in {topic}. Known for your ability\n    to find relevant information and present it clearly.\n\nreporting_analyst:\n  role: \"Reporting Analyst\"\n  goal: \"Create detailed reports based on research data\"\n  backstory: >\n    You're a meticulous analyst who transforms raw data into\n    actionable insights through well-structured reports.\n```\n\n### tasks.yaml\n\n```yaml\nresearch_task:\n  description: >\n    Conduct thorough research about {topic}.\n    Find the most relevant information for {year}.\n  expected_output: >\n    A list with 10 bullet points of the most relevant\n    information about {topic}.\n  agent: researcher\n\nreporting_task:\n  description: >\n    Review the research and create a comprehensive report.\n    Focus on key findings and recommendations.\n  expected_output: >\n    A detailed report in markdown format with executive\n    summary, findings, and recommendations.\n  agent: reporting_analyst\n  output_file: report.md\n```\n\n### crew.py\n\n```python\nfrom crewai import Agent, Crew, Process, Task\nfrom crewai.project import CrewBase, agent, crew, task\nfrom crewai_tools import SerperDevTool\n\n@CrewBase\nclass MyProjectCrew:\n    \"\"\"My Project crew\"\"\"\n\n    @agent\n    def researcher(self) -> Agent:\n        return Agent(\n            config=self.agents_config['researcher'],\n            tools=[SerperDevTool()],\n            verbose=True\n        )\n\n    @agent\n    def reporting_analyst(self) -> Agent:\n        return Agent(\n            config=self.agents_config['reporting_analyst'],\n            verbose=True\n        )\n\n    @task\n    def research_task(self) -> Task:\n        return Task(config=self.tasks_config['research_task'])\n\n    @task\n    def reporting_task(self) -> Task:\n        return Task(\n            config=self.tasks_config['reporting_task'],\n            output_file='report.md'\n        )\n\n    @crew\n    def crew(self) -> Crew:\n        return Crew(\n            agents=self.agents,\n            tasks=self.tasks,\n            process=Process.sequential,\n            verbose=True\n        )\n```\n\n### main.py\n\n```python\nfrom my_project.crew import MyProjectCrew\n\ndef run():\n    inputs = {\n        'topic': 'AI Agents',\n        'year': 2025\n    }\n    MyProjectCrew().crew().kickoff(inputs=inputs)\n\nif __name__ == \"__main__\":\n    run()\n```\n\n## Flows - Event-driven orchestration\n\nFor complex workflows with conditional logic, use Flows:\n\n```python\nfrom crewai.flow.flow import Flow, listen, start, router\nfrom pydantic import BaseModel\n\nclass MyState(BaseModel):\n    confidence: float = 0.0\n\nclass MyFlow(Flow[MyState]):\n    @start()\n    def gather_data(self):\n        return {\"data\": \"collected\"}\n\n    @listen(gather_data)\n    def analyze(self, data):\n        self.state.confidence = 0.85\n        return analysis_crew.kickoff(inputs=data)\n\n    @router(analyze)\n    def decide(self):\n        return \"high\" if self.state.confidence > 0.8 else \"low\"\n\n    @listen(\"high\")\n    def generate_report(self):\n        return report_crew.kickoff()\n\n# Run flow\nflow = MyFlow()\nresult = flow.kickoff()\n```\n\nSee [Flows Guide](references/flows.md) for complete documentation.\n\n## Memory system\n\n```python\n# Enable all memory types\ncrew = Crew(\n    agents=[researcher],\n    tasks=[research_task],\n    memory=True,           # Enable memory\n    embedder={             # Custom embeddings\n        \"provider\": \"openai\",\n        \"config\": {\"model\": \"text-embedding-3-small\"}\n    }\n)\n```\n\n**Memory types:** Short-term (ChromaDB), Long-term (SQLite), Entity (ChromaDB)\n\n## LLM providers\n\n```python\nfrom crewai import LLM\n\nllm = LLM(model=\"gpt-4o\")                              # OpenAI (default)\nllm = LLM(model=\"claude-sonnet-4-5-20250929\")                       # Anthropic\nllm = LLM(model=\"ollama/llama3.1\", base_url=\"http://localhost:11434\")  # Local\nllm = LLM(model=\"azure/gpt-4o\", base_url=\"https://...\")              # Azure\n\nagent = Agent(role=\"Analyst\", goal=\"Analyze data\", llm=llm)\n```\n\n## CrewAI vs alternatives\n\n| Feature | CrewAI | LangChain | LangGraph |\n|---------|--------|-----------|-----------|\n| **Best for** | Multi-agent teams | General LLM apps | Stateful workflows |\n| **Learning curve** | Low | Medium | Higher |\n| **Agent paradigm** | Role-based | Tool-based | Graph-based |\n| **Memory** | Built-in | Plugin-based | Custom |\n\n## Best practices\n\n1. **Clear roles** - Each agent should have a distinct specialty\n2. **YAML config** - Better organization for larger projects\n3. **Enable memory** - Improves context across tasks\n4. **Set max_iter** - Prevent infinite loops (default 15)\n5. **Limit tools** - 3-5 tools per agent max\n6. **Rate limiting** - Set max_rpm to avoid API limits\n\n## Common issues\n\n**Agent stuck in loop:**\n```python\nagent = Agent(\n    role=\"...\",\n    max_iter=10,           # Limit iterations\n    max_rpm=5              # Rate limit\n)\n```\n\n**Task not using context:**\n```python\ntask2 = Task(\n    description=\"...\",\n    context=[task1],       # Explicitly pass context\n    agent=writer\n)\n```\n\n**Memory errors:**\n```python\n# Use environment variable for storage\nimport os\nos.environ[\"CREWAI_STORAGE_DIR\"] = \"./my_storage\"\n```\n\n## References\n\n- **[Flows Guide](references/flows.md)** - Event-driven workflows, state management\n- **[Tools Guide](references/tools.md)** - Built-in tools, custom tools, MCP\n- **[Troubleshooting](references/troubleshooting.md)** - Common issues, debugging\n\n## Resources\n\n- **GitHub**: https://github.com/crewAIInc/crewAI (25k+ stars)\n- **Docs**: https://docs.crewai.com\n- **Tools**: https://github.com/crewAIInc/crewAI-tools\n- **Examples**: https://github.com/crewAIInc/crewAI-examples\n- **Version**: 1.2.0+\n- **License**: MIT","author":"@Orchestra-Research","ownerProfile":null,"authorContacts":null,"sourceUrl":"https://github.com/Orchestra-Research/AI-Research-SKILLs/tree/main/14-agents/crewai","license":"MIT","category":"document","lang":"en","tokens":3049,"stars":0,"calls30d":1,"claimed":false,"visibility":"public","origin":"crawler","version":"0.1.0","createdAt":"2026-08-22","updatedAt":"2026-08-22","files":[{"path":"references/flows.md","size":9846,"sha256":"bb4b104c94ffa8145516d73a1d4af3809723df15d47c2d02b24715bc9a1752f2"},{"path":"references/tools.md","size":10283,"sha256":"fd62a6c2c1ba08fc921cab5dd519c476e38e43f16cbfbba32fd000386ed454f5"},{"path":"references/troubleshooting.md","size":8933,"sha256":"9c55aa955aebe83322bf50e92181ff266f2af9374affbda27dd34e8de79d6fe2"}],"requires":{"mcp":[],"tools":[]},"safety":{"flags":[{"code":"code.eval","kind":"dangerous-code","where":"SKILL.md:231","excerpt":"eval(","message":"evaluates code at runtime","severity":"warn"},{"code":"code.eval","kind":"dangerous-code","where":"references/tools.md:297","excerpt":"eval(","message":"evaluates code at runtime","severity":"warn"}],"scannedAt":"2026-08-22","hasScripts":false,"networkEndpoints":["api.example.com","discord.gg","docs.crewai.com","docs.python.org"]}}