How to Deploy Qwen 3.5‑Plus with CoPaw on Alibaba Cloud ACK/ACS via Agent Sandbox
This guide walks you through deploying the Qwen 3.5‑plus model on Alibaba Cloud ACK/ACS using the ACS Agent Sandbox, creating a CoPaw sandbox, configuring model access, integrating with DingTalk, and optionally using the sandbox’s pause‑and‑wake features.
Background
Qwen 3.5, released by Alibaba in February 2026, is a large‑parameter model that uses a gated‑delta architecture to achieve high performance with only ~5% active compute. CoPaw, an open‑source personal AI assistant built on the AgentScope framework, can be integrated with Qwen 3.5 to provide web‑search, data‑collection, and report‑generation capabilities.
Prerequisites
Create an ACK Pro or ACS cluster that has access to ACS computing power.
Install the kubectl command‑line tool (see the official Kubernetes docs).
When creating the cluster, enable the “Expose API server with EIP” option to allow public access.
Install ACS Agent Sandbox Components
Install the ACK Virtual Node component (skip for ACS clusters).
Install the ack-agent-sandbox-controller component, version >= v0.5.3.
Create the CoPaw Sandbox
Download the temporary KubeConfig from the cluster console and save it to ~/.kube/config (or the Windows equivalent). Then create a copaw.yaml file with the following content:
apiVersion: agents.kruise.io/v1alpha1
kind: Sandbox
metadata:
labels:
app: copaw
name: copaw
namespace: default
spec:
template:
metadata:
labels:
alibabacloud.com/acs: "true"
app: copaw
spec:
containers:
- image: agentscope-registry.ap-southeast-1.cr.aliyuncs.com/agentscope/copaw:latest
imagePullPolicy: IfNotPresent
command: ["copaw", "app"]
name: copaw
ports:
- containerPort: 8088
resources:
limits:
cpu: "2"
memory: 4Gi
requests:
cpu: "2"
memory: 4GiApply the manifest:
$ kubectl create -f copaw.yaml
sandbox.agents.kruise.io/copaw createdAfter creation, a Pod named copaw will start and the service becomes reachable.
Configure CoPaw
Port‑forward the service so the local browser can reach it:
$ kubectl port-forward copaw 8088:8088
Forwarding from 127.0.0.1:8088 -> 8088
Forwarding from [::1]:8088 -> 8088Obtain an API‑Key from the Alibaba Cloud Bailei model service platform (Model Marketplace → API‑Key Management).
Open http://127.0.0.1:8088/models and paste the API‑Key.
Select the desired model ID (e.g., qwen3.5-plus) from the Bailei model list.
Integrate with DingTalk
Follow the DingTalk Open Platform documentation to create an internal application, add the “Robot” capability, and set the message reception mode to “Stream”. After publishing the app, retrieve the AppKey (Client ID) and AppSecret (Client Secret).
In CoPaw’s channel management page ( http://127.0.0.1:8088/channels), add a DingTalk channel and fill in the obtained credentials.
Using CoPaw
Search for the robot name (e.g., rbg-copaw) in DingTalk, start a conversation, and interact with the assistant.
Optional: Sandbox Pause‑and‑Wake
To save resources when CoPaw is idle, edit the sandbox CR to set spec.paused: true. The sandbox will enter a paused state, the Pod disappears, and only minimal storage costs remain.
$ kubectl edit sandbox copaw # set spec.paused to trueResume the sandbox by setting spec.paused: false and applying the change.
$ kubectl edit sandbox copaw # set spec.paused to falseAfter resuming, the sandbox status returns to Running and the Pod is recreated.
Conclusion
Alibaba Cloud ACK combined with the ACS Agent Sandbox provides a ready‑to‑use, secure, and elastic infrastructure for AI agents like CoPaw, enabling rapid deployment, on‑demand scaling, and seamless integration with messaging platforms.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
