Run Agent Spec configurations with OpenAI Agents#
This usage example showcases the creation of a simple Agent Spec Agent, subsequently serialized into JSON and converted into an OpenAI Agents assistant. Also includes mapping of a ServerTool and execution of the conversation.
# Create an Agent Spec agent
from pyagentspec.agent import Agent
from pyagentspec.llms import OpenAiConfig
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_agent = Agent(
name="agentspec_tools_test",
description="agentspec_tools_test",
llm_config=OpenAiConfig(name="gpt-5-mini", model_id="gpt-5-mini"),
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 OpenAI Agents
from agents import Runner, TResponseInputItem
from pyagentspec.adapters.openaiagents import AgentSpecLoader
def subtract(a: float, b: float) -> float:
return a - b
async def main():
loader = AgentSpecLoader(tool_registry={"subtraction-tool": subtract})
assistant = loader.load_json(agentspec_config)
conversation_history: list[TResponseInputItem] = []
while True:
user_input = input("USER >> ")
if user_input == "exit":
break
conversation_history.append({"role": "user", "content": [{"type": "input_text", "text": user_input}]})
result = await Runner.run(assistant, input=[*conversation_history])
conversation_history.extend([item.to_input_item() for item in result.new_items])
print(f"AGENT >> {result.final_output_as(str)}")
# anyio.run(main)
# USER >> Compute 987654321-123456789
# AGENT >> The result of this subtraction is 864197532.