Just Dropped in the AI Playground: New AI Agents, Powerful Customizations, Improved Accessibility

We are thrilled to announce a fresh round of updates to the Stanford AI Playground, designed to make your experience more powerful, customized, accessible, and, well—fun!

For those new to the platform, the Stanford AI Playground is a specialized environment managed by University IT (UIT). The AI Playground allows Stanford faculty, staff, students, postdocs, visiting scholars, and affiliates to more securely experiment with various leading AI models, all in one convenient place.

The latest enhancements, rolled out in early July, focus on improving usability and introducing powerful new tools. Check out the updated look and explore a comparison of models and versions in the Quick Start Guide: 3. Find and explore models. For example, the simplified dropdown for model selection:

New specialized AI agents

The recent AI Playground updates also introduce AI agents. Previously known as plugins, agents can perform tasks like coding, searching documents, or generating images. Agents also offer customization and can integrate a large variety of third-party capabilities and model providers.

We have three AI agents ready to assist with Stanford-specific queries:

• Admin Guide Search: Helps answer questions based on the Stanford Admin Guide.
  • Example prompt: “Are there restrictions on what I can purchase with a Stanford PCard?”
• DoResearch Policy Guide: Allows for AI-assisted searching of Stanford's comprehensive collection of policies and guidelines related to university research activities.
  • Example prompt: “What safety training is required before working in a Stanford lab?”
• Faculty Handbook Search: Designed to help you efficiently find information related to Stanford University faculty policies and guidelines.
  • Example prompt: “What are the guidelines for faculty leave at Stanford?”

Imagen-3, Dall-E 3, Google (web search), ScholarAI, and Wolfram, which were previously available as plugins, now function as AI agents and are listed in the “Agents” group.

Find more details on AI agents in the Quick Start Guide: 10. Try new features.

Temporary chat

Ever wanted to test a quick prompt without cluttering your chat history? The new Temporary Chat feature is the perfect solution. Use it for sensitive topics, quick experiments, or any conversation you don’t need to save permanently. These chats are automatically excluded from your personal search results, cannot be bookmarked, and are deleted after 30 days. Learn more in the Quick Start Guide: 10. Try new features.

Please remember: Do not use high-risk data in your attachments or prompts in the AI Playground (even with Temporary Chat). Visit Responsible AI to learn more about best practices.

“Open Thinking” dropdowns and strengthened Artifacts

For those curious how a model is processing its responses, try out the "Open Thinking" feature with DeepSeek models. You’ll be able to see which criteria and inputs the model is considering as part of its response, allowing you to better tailor any follow-ups. To make sure the Open Thinking content is automatically expanded, simply go to Settings > Chat > Open Thinking Dropdown by Default. Learn more in the Quick Start Guide: 10. Try new features.

Also, to strengthen the Artifacts capability—previewing generated items—be sure to enable shadcn/ui instructions:

1. First go to Settings > Beta Features and set “Toggle Artifacts UI” to on.
2. Also, turn on the toggle below. (Include shadcn/ui components instructions).

Improved accessibility

The Playground also now has improved labels and styling for buttons, more accessible search, improved keyboard focus and screen reader support, and more consistent visual design across the platform. Find more detail in the AI Playground Release Notes.

More help

• To get started, launch the AI Playground.
• For detailed information, use the AI Playground Quick Start Guide and FAQs.
• To share and learn from the Stanford community on the AI Playground, join the #ai-playground Slack channel.
• For technical questions or support, send a Help request.
• To share your feedback with the team, use the feedback form.