Agent Teams

Multi-agent coordination via coordinator and collaborate modes.

Decorator

`team`

Mark a class as a multi-agent team.

Must be called with parentheses:

python

@team(name="research_team", mode="coordinator", model="claude-haiku-4-5")
class ResearchTeam:
    def __init__(
        self,
        researcher: WebResearcherAgent,
        writer: WriterAgent,
    ) -> None: ...

The constructor declares worker agents as typed parameters; they are resolved from the DI container at startup.

Parameters:

Name	Type	Description
`name`	`str`	Human-readable team name (defaults to class name).
`mode`	`Literal['coordinator', 'collaborate']`	"coordinator" routes sub-tasks one at a time; "collaborate" runs all workers sequentially and synthesises.
`model`	`str`	Model for the coordinator LLM calls.
`max_rounds`	`int`	Maximum coordinator-worker rounds before stopping.
`coordinator_prompt`	`str \| None`	Override the default routing prompt.

Metadata

`TeamMeta`

Attached to the class by @team().

Runner & result

`TeamRunner`

Orchestrates a @team() class through its workers.

Usage:

python

runner = TeamRunner(team_cls=ResearchTeam, llm=llm_service)
result = await runner.run("Summarise recent news about AI.")

`TeamResult`

Final result of a team run.

Events

`TeamWorkerStarted`

Emitted when a worker agent begins its sub-task.

`TeamWorkerFinished`

Emitted when a worker agent completes its sub-task.

`TeamCoordinatorDecision`

Emitted when the coordinator decides next action.

`TeamFinalAnswer`

Emitted when the team produces its final answer.