Skip to content

Team Overview

A Team (buddy.team.Team) lets several agents work together on a task under a coordinating leader. The leader is itself driven by a model; it decides how to involve the members and how to assemble the final answer.

Version

Documents buddy-ai 2.2.0. The Team API below is verified against buddy/team/team.py.

Anatomy of a team

from buddy import Agent, Team
from buddy.models.openai import OpenAIChat

researcher = Agent(
    name="Researcher",
    role="Find accurate, up-to-date information",
    model=OpenAIChat(id="gpt-4o-mini"),
)
writer = Agent(
    name="Writer",
    role="Turn research into clear prose",
    model=OpenAIChat(id="gpt-4o"),
)

team = Team(
    members=[researcher, writer],   # REQUIRED, first positional argument
    mode="coordinate",              # default
    model=OpenAIChat(id="gpt-4o"),  # the leader's model
    name="content_team",
    instructions="Research the topic, then write a concise article.",
    markdown=True,
)

team.print_response("Write a short article about quantum computing.")

It's members, not agents

The constructor's first parameter is members — a list of Agent or nested Team objects. There is no agents= argument, no team.collaborate() method, and no TeamController. Give each member a role; give the leader a model.

Coordination modes

The mode parameter selects how the leader engages members:

Mode Behavior
route The leader delegates to the single best member for the task and returns its output.
coordinate (default) The leader delegates to members and synthesizes their outputs into one response.
collaborate Members work together on the task, with the leader facilitating. 
team = Team(members=[researcher, writer], mode="route")

Running a team

Method Returns
team.run(message, ...) A TeamRunResponse (or an event iterator when stream=True)
team.print_response(message, ...) Streams a formatted response to the console
team.arun(message, ...) Async version of run 
response = team.run("Summarize the latest on fusion energy.")
print(response.content)

Useful constructor options

Parameter Default Purpose
name None Human-readable team name
instructions None Guidance for the leader
success_criteria None What "done" looks like
markdown False Ask the leader to format output as Markdown
show_members_responses False Surface each member's response
enable_agentic_context False Let the leader maintain & share team context
share_member_interactions False Pass prior member outputs to members
knowledge None Attach an AgentKnowledge base to the team

Where to go next