Kenhuang Academy
  • English

      • CrewAI Tutorial

      Welcome to the CrewAI tutorial. Based on the official CrewAI docs and the open-source crewAIInc/crewAI project, this guide shows how to ship multi-agent apps with Flows (event-driven backbone) and Crews (role-playing agent teams).

      Official framing: start production apps with a Flow. Put a Crew inside a Flow step only when you need autonomous collaboration.

      Table of Contents

      Basics

      1. Introduction — Positioning, Flows vs Crews, vs LangChain / LlamaIndex
      2. Installation — Python, uv / pip, CLI, coding-agent skills
      3. Quick Start — First Crew, and a minimal Flow with @start / @listen

      Core objects

      1. Agents — role / goal / backstory, tools, kickoff()
      2. Tasks — description, expected_output, context and output
      3. Crews and Processes — Crew, Process.sequential / hierarchical

      Capabilities

      1. Tools — crewai[tools], built-ins, custom BaseTool / @tool
      2. Flows — State, @start / @listen / @router, Crews inside steps
      3. Memory — Unified Memory, four usage patterns

      Practice

      1. Practical Examples — Research/write crew, Flow wrapping a Crew, retrieval tools
      2. Resources — Official docs and related courses on this site

      Learning path

      StageGoalChapters
      Day 1Run a CrewIntro → Install → Quick Start
      Days 2–3Core objectsAgents → Tasks → Crews and Processes
      Week 1Orchestration & memoryTools → Flows → Memory
      Week 2A full projectExamples → Resources

      Prerequisites

      • Python 3.10+ (official: >=3.10 and <3.14)
      • Familiarity with LLM / Chat APIs
      • At least one provider API key (OpenAI, Anthropic, etc.)
      CourseRelationship
      LangChainAgent harness + LangGraph graphs; single-agent / custom graphs
      LlamaIndexIndexing and query engines; CrewAI can wrap them via LlamaIndexTool
      RAGChunking, embeddings, vector stores; this course calls retrieval as tools

      Two ways in

      PathBest forCommand / files
      CLI scaffoldStandard layout, JSONC, later AMP deploycrewai create flow or crewai create crew
      Python scriptsLearning Agent / Task / Crew / Flow firstpip install crewai then kickoff()

      This tutorial covers both. Object semantics follow the Python API; JSONC fields map to the same official Agent / Task / Crew attributes.

      Object cheat sheet

      You want to…Use
      Give a role, goal, and personaAgent(role, goal, backstory)
      Specify work and a done-definitionTask(description, expected_output)
      Assemble a team and a strategyCrew(..., process=Process.sequential|hierarchical)
      Own state, order, and branchesFlow + @start / @listen / @router
      Let agents take actionstools=[...] (crewai[tools] or custom)
      Remember facts across tasksMemory / memory=True

      What you will build

      1. A sequential two-agent Crew — researcher then writer
      2. A stateful Flow — @start sets a topic; a @listen step runs crew.kickoff()
      3. A production-shaped skeleton — search / files / custom tools, memory across tasks

      This course does not cover every AMP enterprise console feature; use the official platform docs for that.

      Start with Introduction.

      评论