QA: Beautiful Chat — PydanticAI

Prerequisites

• Demo is deployed and accessible
• Agent backend is healthy (check /api/health)
• OPENAI_API_KEY set

Test Steps

1. Basic Functionality

• Navigate to /demos/beautiful-chat
• Verify the polished two-column layout renders with the chat column and a toggle between "Chat" and "App" modes
• Verify suggestion pills render in the chat

2. Controlled Generative UI (Charts)

• Click "Pie Chart (Controlled Generative UI)"
• Verify the agent calls query_data then renders a pie chart inline
• Click "Bar Chart" and verify a bar chart renders

3. A2UI

• Click "Search Flights (A2UI Fixed Schema)"
• Verify 2 flight cards render inline via the fixed flight catalog
• Click "Sales Dashboard (A2UI Dynamic)"
• Verify a dashboard with metrics + charts renders

4. Open Generative UI

• Click "Calculator App (Open Generative UI)"
• Verify a sandboxed calculator iframe mounts

5. Shared State (Todos)

• Click "Task Manager (Shared State)"
• Verify the layout flips to "App" mode and todos appear in the right-hand canvas
• Toggle a todo's checkbox; verify the UI updates locally
• Edit a todo title/description; verify it persists

6. Frontend Tools

• Click "Toggle Theme (Frontend Tools)"
• Verify the dark/light theme flips via the toggleTheme tool

7. HITL

• Click "Schedule Meeting (Human In The Loop)"
• Verify a time picker modal opens in chat

Known Limitations vs. langgraph-python port

• MCP Apps (Excalidraw): not wired on the PydanticAI backend; the "Excalidraw Diagram" suggestion pill is intentionally omitted here. Tracked in PARITY_NOTES.md.
• Per-token state streaming: langgraph-python uses StateStreamingMiddleware to stream per-token todo deltas. PydanticAI's agent.to_ag_ui() emits a single STATE_SNAPSHOT on manage_todos completion instead. Functionally equivalent — the todo list still appears — but does not animate character-by-character.

Expected Results

• Fixed-schema flight cards, dynamic-schema dashboards, sandboxed iframes, the HITL time picker, and the shared todos canvas all render in a single combined cell powered by the beautiful_chat PydanticAI agent.