Skip to main content
When you call run() on a team, the leader decides how to handle the request: respond directly, use tools, or delegate to members. Team delegation flow The default flow:
  1. Team receives user input
  2. Leader analyzes the input and decides which members to delegate to
  3. Leader formulates a task for each selected member
  4. Members execute and return results (concurrently in async mode)
  5. Leader synthesizes results into a final response
You can customize this flow with team modes:
ModeConfigurationBehavior
Coordinate (default)mode=TeamMode.coordinate (or omit mode)Leader selects members, formulates tasks, synthesizes results
Routemode=TeamMode.routeLeader routes to one member and returns their response directly
Broadcastmode=TeamMode.broadcastLeader delegates the same task to all members simultaneously
Tasksmode=TeamMode.tasksLeader builds and executes a shared task list until the goal is complete
Use TeamMode from agno.team.mode to set the mode explicitly. The legacy flags still work, but mode is the recommended approach.

​
Coordinate Mode (Default)

The leader controls everything: which members to use, what task to give them, and how to combine their outputs. 
from agno.team import Team
from agno.agent import Agent
from agno.team.mode import TeamMode
from agno.models.openai import OpenAIResponses
from agno.tools.hackernews import HackerNewsTools
from agno.tools.yfinance import YFinanceTools

team = Team(
    name="Research Team",
    model=OpenAIResponses(id="gpt-4o"),
    members=[
        Agent(name="News Agent", role="Get tech news", tools=[HackerNewsTools()]),
        Agent(name="Finance Agent", role="Get stock data", tools=[YFinanceTools()])
    ],
    mode=TeamMode.coordinate,
    instructions="Research the topic thoroughly, then synthesize findings into a clear report."
)

team.print_response("What's happening with AI companies and their stock prices?")
Use this when:
  • Tasks need decomposition into subtasks
  • You want quality control over the final output
  • The leader should add context or reasoning to member outputs

​
Route Mode

The leader selects which member handles the request and returns the member’s response directly. By default the leader can still craft the task; set determine_input_for_members=False to pass the user input through unchanged. Route mode flow 
from agno.team import Team
from agno.agent import Agent
from agno.team.mode import TeamMode
from agno.models.openai import OpenAIResponses

team = Team(
    name="Language Router",
    model=OpenAIResponses(id="gpt-4o"),
    members=[
        Agent(name="English Agent", role="Answer questions in English"),
        Agent(name="Japanese Agent", role="Answer questions in Japanese"),
    ],
    mode=TeamMode.route,
    determine_input_for_members=False # Pass user input unchanged to member
)

team.print_response("How are you?")        # Routes to English Agent
team.print_response("お元気ですか?")        # Routes to Japanese Agent
Use this when:
  • You have specialized agents and want automatic routing
  • The leader shouldn’t modify the request or response
  • You want lower latency (no synthesis step)

​
Legacy Configuration Flags

These flags still work, but are overridden by mode when set. respond_directly=True: Return member responses without leader synthesis (maps to TeamMode.route). Direct response flow determine_input_for_members=False: Send user input directly to members instead of having the leader formulate a task (works with any mode). Raw input flow Combine both for a full passthrough: 
team = Team(
    respond_directly=True,
    determine_input_for_members=False,
    ...
)

​
Broadcast Mode

The leader delegates to all members at once. Useful for gathering multiple perspectives or parallel research. Broadcast mode flow 
import asyncio
from agno.team import Team
from agno.agent import Agent
from agno.team.mode import TeamMode
from agno.models.openai import OpenAIResponses
from agno.tools.hackernews import HackerNewsTools
from agno.tools.arxiv import ArxivTools
from agno.tools.duckduckgo import DuckDuckGoTools

team = Team(
    name="Research Team",
    model=OpenAIResponses(id="gpt-4o"),
    members=[
        Agent(name="HackerNews Researcher", role="Find discussions on HackerNews", tools=[HackerNewsTools()]),
        Agent(name="Academic Researcher", role="Find academic papers", tools=[ArxivTools()]),
        Agent(name="Web Researcher", role="Search the web", tools=[DuckDuckGoTools()]),
    ],
    mode=TeamMode.broadcast,
    instructions="Synthesize findings from all researchers into a comprehensive report."
)

# Use async for concurrent execution
asyncio.run(team.aprint_response("Research the current state of AI agents"))
Use this when:
  • You want multiple perspectives on the same topic
  • Members can work independently
  • Parallel execution improves latency
If both delegate_to_all_members=True and respond_directly=True are set, broadcast wins and respond_directly is disabled.

​
Tasks Mode

Tasks mode is an autonomous loop where the leader decomposes the goal into tasks, executes them, and marks the goal complete. 
from agno.team import Team
from agno.agent import Agent
from agno.team.mode import TeamMode
from agno.models.openai import OpenAIResponses

team = Team(
    name="Ops Team",
    model=OpenAIResponses(id="gpt-4o"),
    members=[
        Agent(name="Research Agent", role="Collect findings"),
        Agent(name="Writer Agent", role="Draft the final report"),
    ],
    mode=TeamMode.tasks,
    max_iterations=6
)

team.print_response("Compile a short report on recent AI agent frameworks.")

​
Structured Input

When using determine_input_for_members=False, you can pass structured Pydantic models directly to members: 
from pydantic import BaseModel, Field
from agno.team import Team
from agno.agent import Agent
from agno.models.openai import OpenAIResponses
from agno.tools.hackernews import HackerNewsTools

class ResearchRequest(BaseModel):
    topic: str
    num_sources: int = Field(default=5)

research_agent = Agent(
    name="Research Agent",
    role="Research topics on HackerNews",
    tools=[HackerNewsTools()]
)

team = Team(
    name="Research Team",
    model=OpenAIResponses(id="gpt-4o"),
    members=[research_agent],
    determine_input_for_members=False  # Pass input directly to member
)

request = ResearchRequest(topic="AI Agents", num_sources=10)
team.print_response(input=request)

​
Production Considerations

​
Token Costs

Each mode has different token overhead:
ModeCoordination CostWhen to Use
CoordinateHigh (decomposition + synthesis)Quality matters more than cost
RouteLow (selection only)Simple routing, cost-sensitive
BroadcastMedium (synthesis only)Parallel research, multiple perspectives
TasksHigh (planning + iterative loop)Multi-step goals with dependencies

​
Latency

  • Coordinate: Sequential (leader thinks → members execute → leader synthesizes)
  • Route: Fast (leader selects → member executes)
  • Broadcast with async: Parallel member execution, but synthesis adds latency
  • Tasks: Iterative; multiple cycles until tasks are complete or max_iterations is reached

​
Error Handling

What happens when a member fails?
  • Coordinate: Leader can work with partial results from other members
  • Route: Failure is returned directly to caller
  • Broadcast: Leader synthesizes available results, may note missing data
  • Tasks: Task list tracks failed/blocked tasks; the leader can retry or reassign until completion

​
Developer Resources