agents.multi.experiments.implementations.multi_agent_v2

Multi-agent V2 with proper state management and rebuilding support.

This module provides a rebuilt MultiAgent that uses MultiAgentState without schema flattening, maintaining type safety and hierarchical access.

Classes

ExecutionMode

Execution modes for multi-agent systems.

MultiAgentV2

Rebuilt multi-agent system using proper state management.

Module Contents

class agents.multi.experiments.implementations.multi_agent_v2.ExecutionMode

Bases: str, enum.Enum

Execution modes for multi-agent systems.

Initialize self. See help(type(self)) for accurate signature.

class agents.multi.experiments.implementations.multi_agent_v2.MultiAgentV2

Bases: haive.agents.base.agent.Agent

Rebuilt multi-agent system using proper state management.

Key improvements: - Uses MultiAgentState without schema flattening - Each agent maintains its own schema - Supports rebuilding with class methods - Proper hierarchical state access

add_agent(agent, rebuild=True)

Add an agent and optionally rebuild.

Parameters:

  • agent (haive.agents.base.agent.Agent) – Agent to add

  • rebuild (bool) – Whether to rebuild the graph

Returns:

Self or new instance if rebuilt

Return type:

MultiAgentV2

build_graph()

Build the multi-agent graph based on execution mode.

Return type:

haive.core.graph.state_graph.base_graph2.BaseGraph

classmethod from_agents(agents, name=None, execution_mode=ExecutionMode.SEQUENCE, **kwargs)

Create MultiAgent from a list or dict of agents.

Parameters:

  • agents (list[haive.agents.base.agent.Agent] | dict[str, haive.agents.base.agent.Agent]) – List or dict of agents to coordinate

  • name (str | None) – Optional name for the multi-agent

  • execution_mode (ExecutionMode) – How to execute agents

  • **kwargs – Additional configuration

Returns:

MultiAgentV2 instance

Return type:

MultiAgentV2

classmethod from_config(config, agents=None)

Create MultiAgent from configuration dict.

Parameters:

  • config (dict[str, Any]) – Configuration dictionary

  • agents (list[haive.agents.base.agent.Agent] | dict[str, haive.agents.base.agent.Agent] | None) – Optional agents (overrides config)

Returns:

MultiAgentV2 instance

Return type:

MultiAgentV2

get_agent(agent_name)

Get an agent by name.

Parameters:

agent_name (str)

Return type:

haive.agents.base.agent.Agent | None

classmethod rebuild_with_agents(original, new_agents, **kwargs)

Rebuild MultiAgent with new agents.

Parameters:

  • original (MultiAgentV2) – Original MultiAgent to rebuild from

  • new_agents (list[haive.agents.base.agent.Agent] | dict[str, haive.agents.base.agent.Agent]) – New agents to use

  • **kwargs – Additional config overrides

Returns:

New MultiAgentV2 instance

Return type:

MultiAgentV2

remove_agent(agent_name, rebuild=True)

Remove an agent and optionally rebuild.

Parameters:

  • agent_name (str) – Name of agent to remove

  • rebuild (bool) – Whether to rebuild the graph

Returns:

Self or new instance if rebuilt

Return type:

MultiAgentV2

setup_agent()

Setup hook - MultiAgent uses MultiAgentState by default.

Return type:

None

setup_multi_agent()

Set up the multi-agent system.

Return type:

MultiAgentV2

classmethod validate_agents(values)

Ensure agents are provided.

Parameters:

values (dict[str, Any])

Return type:

dict[str, Any]