Frequently asked questions¶

Can I use only the UI?¶

Yes. ai-elements-ui depends on ai-elements-chat, not the networking or agent modules. Render a single element, or supply your own ChatBackend.

Do I need the reference server or Pydantic AI?¶

No. They are examples and test infrastructure. A compatible AI SDK / AG-UI / A2A / ACP server, a model API, or your own backend can supply the messages. Start with Choose your setup.

Why does the local endpoint fail?¶

Declare INTERNET; allow cleartext HTTP in the debug manifest only when testing an HTTP server. The emulator reaches your computer through 10.0.2.2, not its own localhost. Check /health, the server bind address and firewall. See App setup.

Why does a tool wait forever for approval?¶

Use Chat(controller) or forward Conversation.onToolDecision to controller.respondToApproval. Also forward onInputResponse for forms. A display-only tool card cannot answer the waiting backend.

Will a ViewModel restore conversations after process death?¶

No. It preserves state across configuration changes. Your app must persist messages and pass them as initialMessages when constructing a new controller after process death.

Why are some modules minSdk 26?¶

Optional integrations and capabilities use newer platform APIs. See the per-artifact installation matrix. The tested build toolchain is separate from runtime minSdk.

Are dependencies all stable releases?¶

No. The current Material 3 Expressive and Compose versions include alpha releases. The exact tested versions are listed in Installation. Our compatibility policy still applies to the library's stable public APIs.

Can I copy a component example into my own app?¶

Yes: add its artifact and the printed imports, wrap it in AiElementsTheme, and provide the state/callbacks named in the function parameters. The same examples compile in an independent Maven consumer build. Screenshots show the Demo's display scenarios, not a separate web implementation.