Swarm & Handoffs

Agent SDK supports multi-agent collaboration, allowing specialized agents to work together.

Agent Bridge

The AgentBridge allows one agent to call another agent as if it were a tool.

Scenario: A "Manager" agent delegates a coding task to a "Coder" agent.

from agent_sdk import Agent, Runner, AgentBridge

# 1. Define Agents
manager = Agent(name="Manager", instructions="Plan the task and delegate technical work.")
coder = Agent(name="Coder", instructions="Write and execute Python code.")

# 2. Bridge Coder to Manager
# This creates a tool named 'ask_coder(task: str)' and gives it to the Manager.
bridge = AgentBridge(agent=coder, runner=runner)
manager.tools["ask_coder"] = bridge.create_tool()

# 3. Run Manager
runner.run_stream(manager, "Create a snake game in Python.")

Flow: 1. Manager receives request. 2. Manager calls ask_coder("Create a snake game..."). 3. Coder receives the message, does the work, and returns the result string. 4. Manager receives the result and reports back to the user.

Agent Swarm

The AgentSwarm creates a mesh network where every agent is connected to every other agent automatically.

from agent_sdk import AgentSwarm

swarm = AgentSwarm(runner)
swarm.add_agent(manager)
swarm.add_agent(coder)
swarm.add_agent(researcher)

# Connects everyone to everyone
swarm.connect_all() 

# Start with any agent
runner.run_stream(manager, "Research the market and build a prototype.")

Handoffs (Transferring Control)

Instead of a sub-call (function call style), an agent can "handoff" the conversation entirely to another agent.

Note: Handoff logic is typically handled by setting a specific return value or flag that the loop detects to switch the active agent_stack.

(Currently, AgentBridge is the primary method for interaction, functioning as a sub-routine call.)