Multi-Agent

Multi-Agent and Composition

Agiwo supports composing agents in two ways: agent-as-tool for lightweight delegation, and the scheduler for persistent orchestration.

Agent-as-Tool

Wrap an agent as a tool that another agent can call:

from agiwo.agent import Agent, AgentConfig
from agiwo.llm import OpenAIModel

researcher = Agent(
    AgentConfig(
        name="researcher",
        description="Researches topics thoroughly and returns summaries",
        system_prompt="You are a research specialist. Be thorough and cite sources.",
    ),
    model=OpenAIModel(id="gpt-4o", name="gpt-4o"),
)

orchestrator = Agent(
    AgentConfig(
        name="orchestrator",
        description="Coordinates research tasks",
        system_prompt="Delegate independent research tasks to the researcher tool.",
    ),
    model=OpenAIModel(id="gpt-4o", name="gpt-4o"),
    tools=[researcher.as_tool()],
)

result = await orchestrator.run("Compare Python and Rust for systems programming")

How it works

1. The orchestrator sees researcher as a regular tool.
2. When called, the researcher agent runs its own execution loop.
3. The researcher’s final response returns as the tool result.
4. The orchestrator continues reasoning with that output.

Scheduler Orchestration

For long-running, persistent multi-agent setups, use the scheduler:

from agiwo.scheduler import Scheduler

async with Scheduler() as scheduler:
    orchestrator_id = await scheduler.submit(
        orchestrator,
        "Coordinate the research pipeline",
        persistent=True,
    )

    await scheduler.wait_for(orchestrator_id)
    await scheduler.enqueue_input(orchestrator_id, "Now analyze the cost implications")

Scheduler Tools

Agents running under the scheduler automatically get orchestration tools such as:

• spawn_agent
• query_spawned_agent
• cancel_agent
• list_agents
• sleep_and_wait

Patterns

Pipeline

Chain agents sequentially when each step depends on the last output.

Fan-out / Fan-in

Spawn multiple agents in parallel, then synthesize their results.

Supervisor Pattern

Keep a persistent supervisor alive and let it create transient workers internally.