Kenhuang Academy
  • English

      • Installation

      Based on the official Installation guide (treat that page as source of truth). CrewAI uses uv for the CLI and project deps; the library also installs with pip install crewai.

      Requirements

      • Python >=3.10 and <3.14
      • uv (official) or pip
      • At least one LLM provider API key
      python3 --version

      Coding-agent skills (optional)

      In Cursor, Claude Code, Codex, and similar:

      npx skills add crewaiinc/skills

      If npx is missing, skip this and use docs.crewai.com plus llms.txt.

      Official path: uv and the CLI

      macOS / Linux:

      curl -LsSf https://astral.sh/uv/install.sh | sh

      Windows:

      powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

      Install the global CLI:

      uv tool install crewai
uv tool update-shell   # if PATH warning
uv tool list           # should list crewai
crewai version
crewai create --help

      Upgrade the CLI with uv tool install crewai --upgrade. That upgrades the global tool only; project venv upgrades are documented under “Upgrading CrewAI in a project”.

      Windows chroma-hnswlib / missing float.h: install Visual Studio Build Tools with Desktop development with C++.

      Alternative: pip

      python -m venv .venv
# Windows: .venv\Scripts\activate
# macOS / Linux: source .venv/bin/activate
pip install -U crewai
pip install 'crewai[tools]'   # crewai_tools: Serper, files, RAG helpers, …
      python -c "from crewai import Agent, Task, Crew; print('ok')"

      In a uv project: uv add crewai or uv add 'crewai[tools]'. Some providers have extras such as uv add "crewai[openai]". Other LiteLLM-backed providers: uv add 'crewai[litellm]' — see LLMs.

      API keys

      Never hardcode secrets. Use .env (gitignored).

      OPENAI_API_KEY=sk-...
MODEL=openai/gpt-4o-mini
SERPER_API_KEY=...
      from dotenv import load_dotenv
load_dotenv()

      Model IDs are provider/model-id (e.g. openai/gpt-4o-mini, anthropic/claude-sonnet-4-6). Or from crewai import LLM → LLM(model="openai/gpt-4o-mini").

      Create a project

      JSON-first Crew (current default):

      crewai create crew my_research_crew
cd my_research_crew
crewai install
crewai run

      Agents live in agents/*.jsonc; tasks, process, and default inputs live in crew.jsonc. {placeholder} values are filled from inputs or CLI prompts.

      Classic Python / YAML (crew.py + config/agents.yaml):

      crewai create crew my_research_crew --classic

      Flow project (the official Quickstart’s production entry):

      crewai create flow latest-ai-flow
cd latest_ai_flow
crewai install
crewai run

      Add packages with uv add <package-name>. CrewAI internal packages may set exclude-newer; your direct dependencies are unaffected.

      Enterprise (awareness)

      • CrewAI AMP (SaaS) — hosted console, including no-code crews
      • CrewAI Factory — self-hosted containers

      This tutorial stays on the local SDK / CLI.

      Troubleshooting

      crewai not found? uv tool update-shell and open a new terminal.

      Python 3.14? Official requirement is currently <3.14; use 3.12 or 3.13.

      Scripts only, no scaffold? pip install crewai then from crewai import Agent, Task, Crew — see Quick Start.

      Next steps

      评论