Agent Spec Adapters - AutoGen

↑ With the Agent Spec adapter for AutoGen, you can easily import agents from external frameworks using Agent Spec and run them with AutoGen.

Microsoft AutoGen supports the development of multi-agent conversational systems, allowing agents to communicate and collaborate to solve tasks.

Get started

To get started, set up your Python environment (Python 3.10 to 3.12 required), and then install the PyAgentSpec package with the AutoGen extension.

python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
pip install "pyagentspec[autogen]"

You are now ready to use the adapter:

• Run Agent Spec configurations with AutoGen (see more details below)

• Convert AutoGen agents to Agent Spec (see more details below)

Run Agent Spec configurations with AutoGen

# Create a Agent Spec agent
from pyagentspec.agent import Agent
from pyagentspec.llms.openaicompatibleconfig import OpenAiCompatibleConfig
from pyagentspec.property import FloatProperty
from pyagentspec.tools import ServerTool

subtraction_tool = ServerTool(
    name="subtraction-tool",
    description="subtract two numbers together",
    inputs=[FloatProperty(title="a"), FloatProperty(title="b")],
    outputs=[FloatProperty(title="difference")],
)

agentspec_llm_config = OpenAiCompatibleConfig(
    name="llama-3.3-70b-instruct",
    model_id="/storage/models/Llama-3.3-70B-Instruct",
    url="url.to.my.llm",
)

agentspec_agent = Agent(
    name="agentspec_tools_test",
    description="agentspec_tools_test",
    llm_config=agentspec_llm_config,
    system_prompt="Perform subtraction with the given tool.",
    tools=[subtraction_tool],
)

# Export the Agent Spec configuration
from pyagentspec.serialization import AgentSpecSerializer

agentspec_config = AgentSpecSerializer().to_json(agentspec_agent)

# Load and run the Agent Spec configuration with AutoGen
from pyagentspec.adapters.autogen import AgentSpecLoader

def subtract(a: float, b: float) -> float:
    return a - b

async def main() -> None:
    converter = AgentSpecLoader(tool_registry={"subtraction-tool": subtract})
    component = converter.load_json(agentspec_config)
    while True:
        input_cmd = input("USER >> ")
        if input_cmd == "q":
            break
        result = await component.run(task=input_cmd)
        print(f"AGENT >> {result.messages[-1].content}")
    await component._model_client.close()

# anyio.run(main)
# USER >> Compute 987654321-123456789
# AGENT >> The result of the subtraction is 864197532.

Convert AutoGen agents to Agent Spec

# Create an AutoGen Agent
import os
os.environ["OPENAI_API_KEY"] = "YOUR_API_KEY"
from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient

async def add_tool(a: int, b: int) -> int:
    """Adds a to b and returns the result"""
    return a + b

autogen_tools = {"add_tool": add_tool}

model_client = OpenAIChatCompletionClient(
    model="gpt-4.1",
)

autogen_agent = AssistantAgent(
    name="assistant",
    model_client=model_client,
    tools=list(autogen_tools.values()),
    system_message="Use tools to solve tasks, and reformulate the answers that you get.",
    reflect_on_tool_use=True,
)

# Convert to Agent Spec
from pyagentspec.adapters.autogen import AgentSpecExporter

agentspec_config = AgentSpecExporter().to_json(autogen_agent)