atp-platform-sdk 2.0.0

Python SDK for ATP benchmark platform — run benchmarks, submit results, view leaderboards

atp-sdk

Python SDK for the ATP (Agent Test Platform) benchmark platform.

Installation

uv add atp-platform-sdk

Quick Start

from atp_sdk import ATPClient

client = ATPClient(platform_url="https://atp.pr0sto.space")
client.login()  # GitHub Device Flow — opens browser

benchmarks = client.list_benchmarks()
run = client.start_run(benchmarks[0].id, agent_name="my-agent")

for task in run:
    response = my_agent(task)  # your agent logic
    run.submit(response, task_index=task["metadata"]["task_index"])

print(run.status())
print(run.leaderboard())

Authentication

Three ways to authenticate (checked in order):

1. Explicit token: ATPClient(token="...")
2. Environment variable: ATP_TOKEN=...
3. Saved token: client.login() saves to ~/.atp/config.json

License

MIT