AI application setup

Build the requested app around its users, data and operations. Keep the existing stack and SDK major unless migration is part of the task; use Bun for new JavaScript projects.

1. Resolve the installed AI SDK major through ai-sdk; use ai-sdk-6 or ai-sdk-7 for implementation. For a new app, default to AI SDK 7: ai@7, @ai-sdk/react@4 and each provider package at latest (provider majors differ; Node.js 22+). Keep v6 only when a required dependency pins it. AI Elements still targets v6, and its context.tsx fails a v7 typecheck until patched (see ai-elements). Read version-matched ai/docs/ and provider docs before using web examples.
2. Use nextjs-shadcn for framework setup and frontend-design for visual direction. Use ai-elements when its components fit the required chat UI; inspect installed source instead of assuming registry props.
3. Keep provider credentials and authorization on the server. Resolve model IDs from configuration or the selected provider's catalog.
4. Implement the smallest complete user flow. Add persistence, approvals, attachments, search, suggestions or dashboards when the app needs them.
5. Match the server stream to the client transport. Verify a second turn, relevant tool states, safe failures and mobile behavior, alongside the project's typecheck/build.

Read for the app being built

• Chatbot: conversation and stream contracts.
• Agent dashboard: execution, approval and run state.
• Examples: current official implementations.
• Project structure: boundaries without a fixed tree.
• nextjs-chatbot: durable history, feedback, embedding and production checks.

These references describe requirements, not copyable production templates.