Source: Crews (CrewAI docs)

CrewAI Crew attributes, kickoff variants, CrewOutput, usage metrics and replay from a task.

Contents

Summary

A Crew binds agents and tasks under a process and adds crew-wide settings: manager LLM, rate limits, memory, tool caching, planning, callbacks and streaming. It documents sync and async kickoff methods and how to replay a run from a given task.

Key claims

  • Process defaults to sequential; hierarchical needs manager_llm or manager_agent.
  • cache (tool results) defaults to true; max_rpm overrides agent-level limits.
  • planning runs a planner before each iteration and adds the plan to task descriptions.
  • akickoff() is native async and recommended for high concurrency; kickoff_async() is a thread wrapper.
  • crewai replay -t <task_id> re-runs from a task, keeping earlier context.

Feeds

Suggest changes →