Crews and kickoff

kickoff is the entry point for every crew run, and its output object holds everything you need afterwards.

A Crew groups agents and tasks and runs them:

crew = Crew(agents=[researcher, writer], tasks=[research, write])
result = crew.kickoff(inputs={"topic": "solar batteries"})

kickoff starts the run. The inputs dictionary fills every {placeholder} in agent and task text. If a placeholder has no matching input, the run fails, so keep names consistent.

By default the crew uses the sequential process: tasks run in the order of the list, one after another (the next lesson looks closer).

kickoff returns a CrewOutput:

result.raw: the final task's answer as text. str(result) gives the same.
result.tasks_output: a list with one TaskOutput per task, useful for debugging what each step produced.
result.token_usage: how many tokens were used, which tells you the cost of the run.
result.pydantic and result.json_dict: structured versions, when you ask for structured output (a later lesson).

Useful crew settings:

verbose=True prints progress for the whole crew.
max_rpm caps model requests per minute across all agents.
output_log_file="crew.log" saves logs to a file.

To run the same crew over many inputs, kickoff_for_each(inputs=[...]) runs it once per dictionary and returns a list of results. Async versions exist too for programs that handle many requests at once.

Example

from crewai import Agent, Crew, Task

writer = Agent(role="Copywriter", goal="Write punchy product copy",
               backstory="You write short, concrete marketing text.")
tagline = Task(description="Write one tagline for {product}.",
               expected_output="A single tagline under 10 words.", agent=writer)

crew = Crew(agents=[writer], tasks=[tagline], verbose=False)

result = crew.kickoff(inputs={"product": "a solar phone charger"})
print(result.raw)                      # Power from the sun, wherever you roam.
print(len(result.tasks_output))        # 1
print(result.token_usage.total_tokens) # e.g. 412

many = crew.kickoff_for_each(inputs=[{"product": "a tent"}, {"product": "a kayak"}])
print([r.raw for r in many])           # two taglines

Remember

Crew(agents, tasks).kickoff(inputs={...}) runs the crew
inputs fill {placeholders} in agents and tasks
CrewOutput has raw, tasks_output and token_usage

Watch out for

Forgetting an input for a placeholder used in a task, which makes kickoff fail before any agent runs.

Check yourself

Where do you find what the second of three tasks produced after kickoff?

Official docs: docs.crewai.com/en/concepts/crews

Example

Remember

Watch out for

Check yourself

Builds on

Unlocks