crewai

name	CrewAI
description	Workflow patterns and gotchas for CrewAI. Directs to RAG for implementation.

CrewAI Workflow

When to Choose CrewAI

• Building role-based multi-agent systems
• Need agents with distinct personas
• Want task delegation patterns
• Building "crew" style collaboration

Decision Framework

Component Selection

Need	Component	RAG Query
Define specialist	Agent	"crewai agent role goal"
Define work unit	Task	"crewai task description"
Coordinate agents	Crew	"crew process sequential"
Custom capabilities	Tool	"crewai custom tool"
Structured output	Output schemas	"crewai output pydantic"

Query RAG: mcp__agentic-rag__query_sdk("component example", sdk="crewai", mode="build")

Critical Gotchas

These define CrewAI success:

1. Role/Goal/Backstory triad - All three shape agent behavior significantly
2. Task descriptions are prompts - Vague descriptions = vague results
3. Expected output matters - Specify format explicitly or get random structure
4. Process type affects flow - sequential vs hierarchical changes everything
5. Tool names must be clear - Agent reads tool name to decide usage
6. Delegation can loop - Agents may delegate back and forth infinitely
7. Memory is per-crew - Not shared across different crew instances

Workflow: Building a CrewAI System

Step 1: Define Agents

RAG Query: mcp__agentic-rag__query_sdk("Agent role goal backstory", sdk="crewai", mode="build")

Each agent needs a clear role, specific goal, and relevant backstory.

Step 2: Create Tools

RAG Query: mcp__agentic-rag__query_sdk("crewai tool decorator", sdk="crewai", mode="build")

Step 3: Define Tasks

RAG Query: mcp__agentic-rag__query_sdk("Task description expected_output", sdk="crewai", mode="build")

Tasks must specify: description, expected_output, assigned agent.

Step 4: Assemble Crew

RAG Query: mcp__agentic-rag__query_sdk("Crew agents tasks process", sdk="crewai", mode="build")

Step 5: Execute

RAG Query: mcp__agentic-rag__query_sdk("crew kickoff inputs", sdk="crewai", mode="build")

Common Error Patterns

Symptom	Likely Cause	RAG Query
Agent off-topic	Bad role/goal	"agent role definition"
Wrong output format	Missing expected_output	"task expected_output"
Infinite delegation	Allow_delegation loop	"delegation control"
Tool not used	Unclear tool name	"tool naming crewai"
Tasks out of order	Wrong process type	"sequential hierarchical"

Process Types

Type	When to Use	RAG Query
Sequential	Tasks depend on previous output	"sequential process"
Hierarchical	Manager delegates to workers	"hierarchical manager"

Crew Patterns

Research Crew

Researcher → Analyst → Writer pipeline. RAG Query: mcp__agentic-rag__query_sdk("research crew example", sdk="crewai", mode="build")

Support Crew

Triage → Specialist routing. RAG Query: mcp__agentic-rag__query_sdk("support crew routing", sdk="crewai", mode="build")

Advanced Features

Query RAG when you need:

• Memory: "crewai memory persistence"
• Callbacks: "crewai task callbacks"
• Async execution: "crewai async kickoff"
• Custom LLM: "crewai custom llm"
• Output validation: "crewai pydantic output"