CrewAI Documentation
by CrewAI
Both halves of CrewAI documented: role-based agent Crews, and the event-driven Flows that make them resumable.
Overview
docs.crewai.com is the reference for CrewAI, the MIT-licensed Python framework for orchestrating role-playing autonomous agents - a project created in October 2023 that now carries roughly 56.5k GitHub stars, 8k forks and daily commits. The docs are organised into Get Started (Introduction, Installation, Quickstart), Build the Basics (Agents, Tasks & Processes, Flows), an Enterprise journey (deploying automations, triggers, team management), and a What's New section covering triggers and integration tools. The framework has two distinct halves and the documentation treats them separately. Crews are teams of agents, each defined by the required role, goal and backstory fields plus optional llm, tools, memory, allow_delegation, max_iter (default 20), respect_context_window for automatic summarisation, and a reasoning flag governed by max_reasoning_attempts. Flows are the event-driven layer, assembled from @start(), @listen() and @router() decorators with or_ and and_ combinators, holding state either as unstructured attributes on self.state or as a typed Pydantic model, and using a @persist decorator that checkpoints to SQLite so long runs can resume under the same ID or fork into a new one. Agents can be declared three ways: JSONC files (the current recommendation), classic config/agents.yaml with @CrewBase decorators, or plain Python for maximum control. The CLI scaffolds real projects via crewai create flow, crewai install and crewai run; the quickstart builds a Serper-backed research crew that writes a markdown report. Installation is uv tool install crewai on Python >=3.10 and <3.14.
At a Glance
- Topic
- Agentic
- Level
- Intermediate
- Format
- Documentation
- Cost
- Free
- Duration
- ~3-4 hours for Get Started plus core concepts, then ongoing reference
- Provider
- CrewAI
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Define agents by role, goal and backstory, then attach tools and an LLM
- ✓Compose agents into a crew and pick the right process for the task
- ✓Build event-driven Flows using the @start, @listen and @router decorators
- ✓Manage flow state as unstructured attributes or as typed Pydantic models
- ✓Persist, resume and fork long-running workflows with the @persist SQLite checkpoint
- ✓Scaffold and execute a project with crewai create flow, install and run
- ✓Configure agents three ways - JSONC files, classic YAML, or directly in Python
- ✓Contain runaway loops using max_iter, reasoning attempts and context-window handling
Highlights
- •Documents both halves of the framework - autonomous Crews and deterministic Flows - not just the demo path
- •The quickstart generates a working scaffolded project with real file layout, not a copy-paste snippet
- •Covers state persistence, resume and fork, which most agent-framework docs skip entirely
- •An explicit telemetry page names CREWAI_DISABLE_TELEMETRY and OTEL_SDK_DISABLED so collection can be switched off
- •Tracks a ~56.5k-star MIT-licensed repo pushed daily, so examples match the shipped API
- •Flags the current JSONC configuration format while keeping the legacy YAML path documented for existing projects
Who It’s For
Best For
- ✓Python developers building their first multi-agent application
- ✓Teams prototyping agent workflows who want scaffolding over framework plumbing
- ✓Engineers evaluating CrewAI against LangGraph, AutoGen or plain function calling
- ✓Developers who need auditable, resumable agent runs rather than one-shot scripts
Prerequisites
- •Python 3.10 or newer but below 3.14 - the docs pin this exact range
- •An LLM provider API key; agents default to a GPT-4-class model
- •Comfort with decorators, type hints and Pydantic models for the Flows sections
- •No prior agent-framework experience required - the quickstart starts from installation
FAQ
What is CrewAI Documentation?
The official documentation for CrewAI, the MIT-licensed Python framework for building multi-agent systems. It covers both of the framework's halves: Crews, where role-based agents collaborate on tasks, and Flows, the event-driven layer that adds routing, typed state and persistence. Working through Get Started and the core concepts leaves you able to scaffold a project, wire agents to tools, and ship an agent workflow that can resume after a crash.
Is CrewAI Documentation free?
CrewAI Documentation is free to access.
What level is CrewAI Documentation for?
CrewAI Documentation is aimed at a intermediate audience. Recommended background: Python 3.10 or newer but below 3.14 - the docs pin this exact range, An LLM provider API key; agents default to a GPT-4-class model, Comfort with decorators, type hints and Pydantic models for the Flows sections, No prior agent-framework experience required - the quickstart starts from installation.
How long does CrewAI Documentation take?
Expect roughly ~3-4 hours for Get Started plus core concepts, then ongoing reference. Most learners work through it at their own pace.
What will I learn from CrewAI Documentation?
You'll learn: Define agents by role, goal and backstory, then attach tools and an LLM; Compose agents into a crew and pick the right process for the task; Build event-driven Flows using the @start, @listen and @router decorators; Manage flow state as unstructured attributes or as typed Pydantic models; Persist, resume and fork long-running workflows with the @persist SQLite checkpoint; Scaffold and execute a project with crewai create flow, install and run; Configure agents three ways - JSONC files, classic YAML, or directly in Python; Contain runaway loops using max_iter, reasoning attempts and context-window handling.
Topics
Sources
This page was written from 6 sources, 2 on domains other than docs.crewai.com.