How to Deploy OpenClaw AI Agent on Alibaba Cloud PAI‑DSW with One Click

OpenClaw Overview

OpenClaw is an AI agent framework that can directly control the operating system, maintain persistent memory, schedule proactive pushes, and interact via Web UI, DingTalk, and other channels. It includes a rich ecosystem of Skills.

Why Deploy on Alibaba Cloud PAI‑DSW

Instant use without operations

Elastic compute, pay‑as‑you‑go

Seamless integration with Alibaba Cloud services

Secure access, no exposed services

Agent runs next to GPU compute

One‑Click Deployment Steps

Step 1 – Open a DSW instance

In the PAI‑DSW console, select the created instance and click “Open”.

Step 2 – Run the installer script

curl -fsSL https://pai-dsw-ai-machine.oss-cn-beijing.aliyuncs.com/agent/openclaw/openclaw_installer_dsw.sh -o openclaw_installer_dsw.sh && bash openclaw_installer_dsw.sh

The script guides you through configuration; choose the “recommended version”. Installation usually takes 2–5 minutes.

Step 3 – Configure the model service

Select either Alibaba Cloud DashScope (百炼) or PAI EAS as the large‑model backend.

◆ 配置 AI 模型
  [1] ○ DashScope – Alibaba Cloud model service
  [2] ○ EAS – PAI EAS service

For DashScope set the Base URL to https://dashscope.aliyuncs.com/compatible-mode/v1 and provide your API Key. For EAS obtain the endpoint and token from the PAI‑EAS console and enter them when prompted.

Step 4 – Tool‑call support

The installer asks whether to disable tool calls. If the EAS service already supports --enable-auto-tool-choice, answer “No”; otherwise answer “Yes” to avoid 400 errors.

Step 5 – Finish configuration

After the script finishes, the OpenClaw gateway starts automatically and prints a gateway URL such as http://127.0.0.1:18789/?token=xxxx. Clicking the link (or copying it into a browser) opens the OpenClaw Web UI. The address is internal to the DSW instance; the gateway proxies the request securely for the logged‑in Alibaba Cloud user.

Access Methods

DSW Gateway Web UI (default)

Optional DingTalk bot integration

SSH tunnel to the local gateway

Typical Use Cases

Daily Q&A and code generation

请帮我在/mnt/workspace下创建一个使用kaggle公开数据实现房价预测的ipynb代码

Intelligent file search

请帮我找一下DSW实例中所有包含"报告"关键词的 Markdown 文件

Scheduled news subscription

每天早上9点，给我发送最新的AI行业新闻

GPU training monitoring

openclaw cron add \
  --name "GPU训练监控" \
  --cron "*/15 * * * *" \
  --tz "Asia/Shanghai" \
  --session isolated \
  --message "请执行以下操作检查GPU训练状态：
1. 运行命令：python gpu_monitor.py
2. 如果返回 'HEARTBEAT_OK'，说明一切正常，无需进一步操作
3. 如果返回异常报告，请：
- 通过钉钉发送完整报告给我
- 分析报告中提到的可能原因
- 给出建议的解决方案
注意：只有检测到异常时才需要通知我。" \
  --announce

Automatic experiment log archiving

请每天晚上10点执行实验日志归档任务：
1. 运行归档脚本： python /gpu-training-tester/experiment_archiver.py
2. 如果返回 'HEARTBEAT_OK'，说明今天没有新的实验，无需操作
3. 如果返回实验记录，请：
   - 将记录追加写入 /mnt/workspace/experiment_log.md
   - 使用 memory 功能记住这些实验结果
   - 发送摘要到钉钉
后续查询历史实验时，OpenClaw 能从记忆中检索信息。

Verification

When the gateway starts, the terminal displays service information (PID, log file, PID file) and the access URL. Opening the URL shows the Web UI with a status bar; “Health: OK” and “Gateway: Connected” indicate a healthy service.

Resources

OpenClaw GitHub repository: https://github.com/openclaw/openclaw

Official documentation: https://docs.openclaw.ai/

Popular Skills collection: https://github.com/VoltAgent/awesome-openclaw-skills