Use AgentBox, Data Science Workshop (DSW)'s built-in AI coding toolkit, with Qwen Code, OpenCode, Claude Code, CodeX, Hermes, and OpenClaw. PAI Token Service provides model access without requiring an API key.
Product highlights
-
Integrated tools: Use Qwen Code, OpenCode, Claude Code, CodeX, Hermes, and OpenClaw.
-
Set up without API keys: Configure model services with PAI Token Service without requesting or managing API keys.
-
Updated Terminal: Use a file tree, file previews, Markdown rendering, and custom font sizes for AI coding.
-
Ready to use: Start without installing dependencies or configuring an environment.
AI coding tools
DSW includes the following AI coding tools for coding, code completion, code-related Q&A, and agent-based tasks.
-
Qwen Code
An AI coding assistant developed by Alibaba Cloud. It supports code completion, code-related Q&A, code explanations, and unit test generation with Qwen models.
Use cases: Everyday coding and completion, code-related Q&A, and code review and optimization.
-
OpenCode
An open-source AI coding tool that supports multiple models and interactive coding. It suits developers who prefer flexible configuration.
Use cases: Comparing models, interactive code development, and rapid prototyping.
-
Claude Code
Anthropic's command-line interface (CLI) AI coding assistant. Use it to write code, run tests, and manage Git from a local terminal. Use cases: Terminal-based development, complex refactoring across files, and automated bug fixes.
-
CodeX
An AI code-generation system developed by OpenAI. It converts natural-language prompts into high-quality code and supports multiple popular languages. Use cases: Converting natural language to code, rapidly developing scripts, and assisting with API calls.
General-purpose agent tools
DSW also includes the following general-purpose agent tools:
-
Hermes
An AI coding agent that independently plans and executes complex, multistep development tasks.
Use cases: Breaking down and executing complex requirements, coordinated changes across files, and end-to-end development tasks.
-
OpenClaw
An AI agent framework that can control the operating system, maintain persistent memory, and manage scheduled tasks. It supports a web UI and DingTalk interactions and includes a broad skills ecosystem.
Use cases: Monitoring and alerting for training tasks, automated experiment-data archiving, and custom agent workflows.
Quick start
Step 1: Create or restart a DSW instance
-
New instance: Sign in to the PAI-DSW console and create an instance.
-
Existing instance (started on or before June 2026): In the instance list, click Restart for the target instance. The new AI development tools are available after the restart.
Step 2: Configure a model service
AI coding tools require a connection to a foundation model. Use PAI Token Service to access a model service without deploying a model or managing API keys.
You can also configure API keys for Model Studio, Elastic Algorithm Service (EAS), or another OpenAI-compatible model service.
Enable PAI Token Service
-
Open the PAI Token Service console.
-
If the service isn't enabled, click Activate to enable it.
-
In DSW, select an AI coding tool, click the PAI Token Service configuration entry, and complete the setup.
Recommended models
Choose a model based on your coding task:
|
Scenario |
Recommended model |
Description |
|
General coding and refactoring |
|
Flagship foundation model with the strongest overall capabilities. Default recommendation. |
|
Coding tasks |
|
Tuned for code generation and long-context workloads. |
|
Complex reasoning and large-scale refactoring |
|
Strong reasoning and planning for architecture-level tasks. |
|
Fast response times |
|
Lightweight and fast for simple questions, renaming, and comment generation. |
|
Cost-effective everyday use |
|
Supports daily conversations and minor changes for common development tasks. |
|
GLM series |
|
Zhipu GLM models. Choose based on your preference. |
Step 3: Start coding
Open the DSW instance. Select an AI coding tool and a model—for example, OpenCode with qwen3.8-max—and start coding.
Updated Terminal
DSW Terminal includes enhancements for AI coding.
-
File tree and management
The file tree panel on the left displays the working-directory structure. Create, rename, and delete files without switching applications.
-
Text file preview
Click a text file in the file tree to preview it with syntax highlighting, without opening another editor.
-
Markdown rendering
Terminal renders
.mdfiles as rich text so you can review their layout. -
Custom font size
Click AA in the lower-left corner of Terminal to adjust the font size.
Usability improvements
-
Pin recently used apps
The DSW Launcher automatically pins recently used apps for quick access.
-
Switch light and dark modes
Click the
button in the upper-right corner of DSW to switch between Light and Dark modes for different lighting conditions.
FAQ
Q: My instance UI doesn't match the documentation. Where are the AI tools?
For an existing instance, restart it and try again. If a new instance still doesn't show the tools, contact PAI technical support.
Q: Does the PAI Token Service cost extra?
PAI Token Service is billed based on actual token use. Charges appear on your PAI Token Service bill.
Q: Can I use my own API key?
Yes. In addition to PAI Token Service, you can configure API keys for Model Studio, Elastic Algorithm Service (EAS), or another OpenAI-compatible model service.
Q: Are there instance specification requirements for AI coding tools?
No additional instance type requirements apply. For smooth performance, choose an instance type with at least 4 vCPUs. If GPU-accelerated inference is required, choose a GPU-enabled instance type.
Q: Will restarting an older instance remove its configuration?
If a cloud disk or cloud storage system, such as Apsara File Storage NAS (NAS) or Cloud Parallel File Storage (CPFS), is mounted in the working directory, restarting updates only DSW components. Code, data, and existing configuration in the instance remain unchanged.