crewai-multi-agent
Multi-agent orchestration framework for autonomous AI collaboration. Use when building teams of specialized agents working together on complex tasks, when you need role-based agent collaboration with memory, or for production workflows requiring sequential/hierarchical execution. Built without LangChain dependencies for lean, fast execution.
- License 许可
- MIT
- 最近更新
- 2026-11-08
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- crewai-multi-agent
- license
- MIT
- metadata
- { "tags": "Agents, CrewAI, Multi-Agent, Orchestration, Collaboration, Role-Based, Autonomous, Workflows, Memory, Production", "version": "1.0.0", "category": "llm-applications", "maintainer": "Kalaris Labs", "dependencies": "crewai>=1.2.0, crewai-tools>=1.2.0" }
- description
- Multi-agent orchestration framework for autonomous AI collaboration. Use when building teams of specialized agents working together on complex tasks, when you need role-based agent collaboration with memory, or for production workflows requiring sequential/hierarchical execution. Built without LangChain dependencies for lean, fast execution.
02 资源文件 5
- 参考文档references/flows.md9.6 KB
- 参考文档references/tools.md9.9 KB
- 参考文档references/troubleshooting.md8.7 KB
- 参考文档references/yaml-configuration-recommended.md2.8 KB
- 脚本scripts/safe_arithmetic.py1.3 KB
03 SKILL.md 正文
CrewAI - Multi-Agent Orchestration Framework
Build teams of autonomous AI agents that collaborate to solve complex tasks.
When to use CrewAI
Use CrewAI when:
- Building multi-agent systems with specialized roles
- Need autonomous collaboration between agents
- Want role-based task delegation (researcher, writer, analyst)
- Require sequential or hierarchical process execution
- Building production workflows with memory and observability
- Need simpler setup than LangChain/LangGraph
Key features:
- Standalone: No LangChain dependencies, lean footprint
- Role-based: Agents have roles, goals, and backstories
- Dual paradigm: Crews (autonomous) + Flows (event-driven)
- 50+ tools: Web scraping, search, databases, AI services
- Memory: Short-term, long-term, and entity memory
- Production-ready: Tracing, enterprise features
Use alternatives instead:
- LangChain: General-purpose LLM apps, RAG pipelines
- LangGraph: Complex stateful workflows with cycles
- AutoGen: Microsoft ecosystem, multi-agent conversations
- LlamaIndex: Document Q&A, knowledge retrieval
Quick start
Installation
# Core framework
pip install crewai
# With 50+ built-in tools
pip install 'crewai[tools]'
Create project with CLI
# Create new crew project
crewai create crew my_project
cd my_project
# Install dependencies
crewai install
# Run the crew
crewai run
Simple crew (code-only)
from crewai import Agent, Task, Crew, Process
# 1. Define agents
researcher = Agent(
role="Senior Research Analyst",
goal="Discover cutting-edge developments in AI",
backstory="You are an expert analyst with a keen eye for emerging trends.",
verbose=True
)
writer = Agent(
role="Technical Writer",
goal="Create clear, engaging content about technical topics",
backstory="You excel at explaining complex concepts to general audiences.",
verbose=True
)
# 2. Define tasks
research_task = Task(
description="Research the latest developments in {topic}. Find 5 key trends.",
expected_output="A detailed report with 5 bullet points on key trends.",
agent=researcher
)
write_task = Task(
description="Write a blog post based on the research findings.",
expected_output="A 500-word blog post in markdown format.",
agent=writer,
context=[research_task] # Uses research output
)
# 3. Create and run crew
crew = Crew(
agents=[researcher, writer],
tasks=[research_task, write_task],
process=Process.sequential, # Tasks run in order
verbose=True
)
# 4. Execute
result = crew.kickoff(inputs={"topic": "AI Agents"})
print(result.raw)
Core concepts
Agents - Autonomous workers
from crewai import Agent
agent = Agent(
role="Data Scientist", # Job title/role
goal="Analyze data to find insights", # What they aim to achieve
backstory="PhD in statistics...", # Background context
llm="gpt-4o", # LLM to use
tools=[], # Tools available
memory=True, # Enable memory
verbose=True, # Show reasoning
allow_delegation=True, # Can delegate to others
max_iter=15, # Max reasoning iterations
max_rpm=10 # Rate limit
)
Tasks - Units of work
from crewai import Task
task = Task(
description="Analyze the sales data for Q4 2024. {context}",
expected_output="A summary report with key metrics and trends.",
agent=analyst, # Assigned agent
context=[previous_task], # Input from other tasks
output_file="report.md", # Save to file
async_execution=False, # Run synchronously
human_input=False # No human approval needed
)
Crews - Teams of agents
from crewai import Crew, Process
crew = Crew(
agents=[researcher, writer, editor], # Team members
tasks=[research, write, edit], # Tasks to complete
process=Process.sequential, # Or Process.hierarchical
verbose=True,
memory=True, # Enable crew memory
cache=True, # Cache tool results
max_rpm=10, # Rate limit
share_crew=False # Opt-in telemetry
)
# Execute with inputs
result = crew.kickoff(inputs={"topic": "AI trends"})
# Access results
print(result.raw) # Final output
print(result.tasks_output) # All task outputs
print(result.token_usage) # Token consumption
Process types
Sequential (default)
Tasks execute in order, each agent completing their task before the next:
crew = Crew(
agents=[researcher, writer],
tasks=[research_task, write_task],
process=Process.sequential # Task 1 → Task 2 → Task 3
)
Hierarchical
Auto-creates a manager agent that delegates and coordinates:
crew = Crew(
agents=[researcher, writer, analyst],
tasks=[research_task, write_task, analyze_task],
process=Process.hierarchical, # Manager delegates tasks
manager_llm="gpt-4o" # LLM for manager
)
Using tools
Built-in tools (50+)
pip install 'crewai[tools]'
from crewai_tools import (
SerperDevTool, # Web search
ScrapeWebsiteTool, # Web scraping
FileReadTool, # Read files
PDFSearchTool, # Search PDFs
WebsiteSearchTool, # Search websites
CodeDocsSearchTool, # Search code docs
YoutubeVideoSearchTool, # Search YouTube
)
# Assign tools to agent
researcher = Agent(
role="Researcher",
goal="Find accurate information",
backstory="Expert at finding data online.",
tools=[SerperDevTool(), ScrapeWebsiteTool()]
)
Custom tools
from crewai.tools import BaseTool
from pydantic import Field
from scripts.safe_arithmetic import calculate
class CalculatorTool(BaseTool):
name: str = "Calculator"
description: str = "Performs mathematical calculations. Input: expression"
def _run(self, expression: str) -> str:
try:
result = calculate(expression)
return f"Result: {result}"
except Exception as e:
return f"Error: {str(e)}"
# Use custom tool
agent = Agent(
role="Analyst",
goal="Perform calculations",
tools=[CalculatorTool()]
)
YAML configuration (recommended)
Details, code examples and parameter tables: references/yaml-configuration-recommended.md. Read it when this step applies.
Flows - Event-driven orchestration
For complex workflows with conditional logic, use Flows:
from crewai.flow.flow import Flow, listen, start, router
from pydantic import BaseModel
class MyState(BaseModel):
confidence: float = 0.0
class MyFlow(Flow[MyState]):
@start()
def gather_data(self):
return {"data": "collected"}
@listen(gather_data)
def analyze(self, data):
self.state.confidence = 0.85
return analysis_crew.kickoff(inputs=data)
@router(analyze)
def decide(self):
return "high" if self.state.confidence > 0.8 else "low"
@listen("high")
def generate_report(self):
return report_crew.kickoff()
# Run flow
flow = MyFlow()
result = flow.kickoff()
See Flows Guide for complete documentation.
Memory system
# Enable all memory types
crew = Crew(
agents=[researcher],
tasks=[research_task],
memory=True, # Enable memory
embedder={ # Custom embeddings
"provider": "openai",
"config": {"model": "text-embedding-3-small"}
}
)
Memory types: Short-term (ChromaDB), Long-term (SQLite), Entity (ChromaDB)
LLM providers
from crewai import LLM
llm = LLM(model="gpt-4o") # OpenAI (default)
llm = LLM(model="claude-sonnet-4-5-20250929") # Anthropic
llm = LLM(model="ollama/llama3.1", base_url="http://localhost:11434") # Local
llm = LLM(model="azure/gpt-4o", base_url="https://...") # Azure
agent = Agent(role="Analyst", goal="Analyze data", llm=llm)
CrewAI vs alternatives
| Feature | CrewAI | LangChain | LangGraph |
|---|---|---|---|
| Best for | Multi-agent teams | General LLM apps | Stateful workflows |
| Learning curve | Low | Medium | Higher |
| Agent paradigm | Role-based | Tool-based | Graph-based |
| Memory | Built-in | Plugin-based | Custom |
Best practices
- Clear roles - Each agent should have a distinct specialty
- YAML config - Better organization for larger projects
- Enable memory - Improves context across tasks
- Set max_iter - Prevent infinite loops (default 15)
- Limit tools - 3-5 tools per agent max
- Rate limiting - Set max_rpm to avoid API limits
Common issues
Agent stuck in loop:
agent = Agent(
role="...",
max_iter=10, # Limit iterations
max_rpm=5 # Rate limit
)
Task not using context:
task2 = Task(
description="...",
context=[task1], # Explicitly pass context
agent=writer
)
Memory errors:
# Use environment variable for storage
import os
os.environ["CREWAI_STORAGE_DIR"] = "./my_storage"
References
- Flows Guide - Event-driven workflows, state management
- Tools Guide - Built-in tools, custom tools, MCP
- Troubleshooting - Common issues, debugging
Resources
- GitHub: https://github.com/crewAIInc/crewAI
- Docs: https://docs.crewai.com
- Tools: https://github.com/crewAIInc/crewAI-tools
- Examples: https://github.com/crewAIInc/crewAI-examples
- Version: 1.2.0+
- License: MIT
Agent operating procedure
- Check the environment. Confirm the framework version, model provider, API keys and rate limits.
- Pin down the inputs. Confirm formats, identifiers and parameters from the data or the user. Ask rather than guess any value that changes the result.
- Run a small version first. Test a single call or chain with a known input and inspect raw outputs.
- Execute the full task using the instructions and references above.
- Validate the result. Evaluate on a small labeled set; check structured outputs against their schema; log prompts and responses.
- Report. State what was run (versions, commands, parameters), what was checked, and what is still uncertain.
| If this happens | Do this |
|---|---|
| Outputs do not match the expected schema | Add validation and retries, tighten the schema, or simplify the prompt. |
| A function, flag or endpoint in these instructions is missing in the installed version | Check the installed version's own documentation (help(), --help, official docs), adapt, and tell the user. Never invent an API. |
| A required input, identifier or parameter is ambiguous | Ask the user, or state the assumption explicitly before running. |
Integrity rules
- Never fabricate results, parameters, identifiers, citations or statistics. If something cannot be run or verified, say so plainly.
- Never send private or sensitive data to external APIs without the user's consent.
- Treat version-specific details here as possibly outdated: confirm them against the official documentation for the installed version.
- Ask before actions that cost money, consume shared GPUs or cloud quota, touch personal or patient data, or cannot be undone.
Related skills
autogpt-agents: Autonomous AI agent platform for building and deploying continuous agents.langchain: Framework for building LLM-powered applications with agents, chains, and RAG.aeon: This skill should be used for time series machine learning tasks including classification, regression, clustering, forecasting, anomaly det…
04 同仓库更多 Skills
youtube-full
Use when the user needs YouTube transcripts, video search, channel browsing, playlist extraction, or content monitoring. Trigger phrases: 'get the transcript for', 'search YouTube for', 'what are the latest videos on', 'list this playlist', 'monitor this channel', or any request involving a YouTube URL, video ID, or @handle. Do NOT use for downloading video or audio files, YouTube engagement data (likes, comments), or private/age-restricted videos.
x-twitter-growth
X/Twitter growth engine for building audience, crafting viral content, and analyzing engagement. Use when the user wants to grow on X/Twitter, write tweets or threads, analyze their X profile, research competitors on X, plan a posting strategy, or optimize engagement. Complements social-content (generic multi-platform) with X-specific depth: algorithm mechanics, thread engineering, reply strategy, profile optimization, and competitive intelligence via web search.
write-a-skill
Create new agent skills with proper structure, progressive disclosure, and bundled resources. Use when user wants to create, write, build, or author a new skill.
wrap-up
Close out a launched Claude Managed Agent — recap every primitive the founder now owns, regenerate the single-file overview page, and suggest the next 1-2 upgrades. Use when the user says "wrap up", "close this out", "what do I own now", "give me the summary", "recap the agent", or when the orchestrator routes phase=wrap-up. primitives_inventory.py tables everything owned (agent, environment, session, memory, outcome, deployment); overview_page.py regenerates a self-contained ./my-agent/agent-overview.html; upgrade_suggester.py ranks the next moves from recorded deferrals plus standing hardening steps. Companion to run-without-you; the last stop before phase=done.
workflow-builder
Design and write deterministic multi-agent workflow scripts (.js files in .claude/workflows/) for Claude Code's Workflow tool. Use when a user wants to build, create, author, scaffold, or run a custom Claude Code workflow, orchestrate sub-agents (fan-out, pipeline, loop, judge-panel), or automate a repeatable multi-step task across fresh-context agents.
weekly-review
Use when someone wants to run a weekly review, close open loops, audit stalled projects and commitments, get their system back to trusted, restart a lapsed review habit, or says "/cs:weekly-review". Walks David Allen's three-phase loop — GET CLEAR, GET CURRENT, GET CREATIVE — with deterministic scripts that inventory open loops, gate the checklist with named gaps, and score commitment health 0-100.