ariacompute-agent

The Aria agent SDK for Python. The API mirrors the OpenAI Agents SDK (Agent, Runner.run, Runner.run_streamed, function_tool, Session) and talks to an aria-agent-cloud deployment through its beta Agents REST API.

pip install ariacompute-agent

import asyncio

from ariacompute_agent import Agent, Runner


async def main() -> None:
    agent = Agent(
        name="History tutor",
        instructions="Answer history questions clearly and concisely.",
        model="gpt-4o-mini",
    )
    result = await Runner.run(agent, "When did the Roman Empire fall?")
    print(result.final_output)


asyncio.run(main())

Tools

from ariacompute_agent import function_tool


@function_tool
def history_fun_fact() -> str:
    """Return a short history fact."""
    return "Sharks are older than trees."


agent = Agent(name="History tutor", tools=[history_fun_fact])

The cloud executes tools inside its own sandbox, so only the schema (name / description / parameters) is sent to the server.

Streaming

streamed = await Runner.run_streamed(agent, "Tell me something surprising")
async for ev in streamed.events:
    if ev["type"] == "agent.turn.output_text.delta":
        print(ev["delta"], end="")
result = await streamed.completed
print(result.final_output)

Frames use the agent.* envelope; agent.turn.completed (or agent.turn.failed) is the only terminal event — there is no data: [DONE] sentinel.

Sessions

session = Session.create(agent)
await Runner.run(agent, "hello", session=session)
await Runner.run(agent, "and then?", session=session)

Configuration

Tests

python3 -m unittest discover -s tests