DuanFlow

DuanFlow is a Python SDK and CLI for submitting Python functions and existing repository commands to DuanFlow cloud GPU/NPU compute.

It is designed for developers and coding agents that need GPU/NPU acceleration, isolated execution, long-running jobs, structured logs, and artifacts.

Install

pip install duanflow

Login

Use browser/link login. Do not paste passwords, API keys, or tokens into an agent chat.

duanflow auth login --api-base https://duanflow.zeabur.app
duanflow whoami

Run a Cloud Function

import subprocess
import duanflow as df

app = df.App("repo-eval")

@app.function(gpu="L40S", timeout=3600, artifacts=["reports/"])
def run_eval():
    subprocess.run(
        "pytest evals/ --json-report --json-report-file reports/eval.json",
        shell=True,
        check=True,
    )
    return {"report": "reports/eval.json"}

print(run_eval.remote(_duanflow_wait=True, _duanflow_stream_logs=True))

Agent Usage

Agents should first decide whether a task can benefit from GPU/NPU or cloud execution. Good fits include model training, inference, evals, embeddings, reranking, image/audio/video processing, simulations, tensor-heavy workloads, and large data processing that can use CuPy, cuDF, RAPIDS, PyTorch, JAX, TensorFlow, torch-npu, or vendor NPU runtimes.

Before submitting a billable cloud job, agents should:

1. Explain the acceleration or isolation reason.
2. Run duanflow whoami.
3. Estimate runtime and cost.
4. Check balance or organization credit.
5. Ask for explicit confirmation.
6. Report job id, logs, cost, and artifacts after completion.

Documentation

• Docs: https://duanflow.zeabur.app/docs.html
• Examples: https://duanflow.zeabur.app/examples.html
• Hosted agent skill: https://duanflow.zeabur.app/skills/duanflow-cloud-exec/SKILL.md