Try whoopsie. Watch your trace land live.
A real chat, calling claude-haiku-4-5 through the Vercel AI Gateway, wrapped in @pisama/sdk in metadata-only mode. Every visitor adds activity to the same public dashboard at /live/ws_demo_public. The buttons below it post canned failure events so you can see what each detector looks like without burning model tokens.
chat
Ask anything. Your prompt and the model's response are redacted to metadata-only — only token counts, finish reasons, and detector verdicts hit the dashboard below. The text never leaves your browser session.
trigger a specific failure
These don't call the model — they post canned trace events to /api/v1/spans so you can watch each detector fire on the live tail without burning tokens.
live tail
whoopsieconnecting…
project: ws_demo_publicWaiting for your first event…
Open your app and send a chat message. The first call will land here within a second.
WHOOPSIE_PROJECT_ID=ws_demo_public
haven't installed yet? get the prompt →Open in a separate tab if you want a bigger view: /live/ws_demo_public
Want this in your own app?
The chat above is one wrapped streamTextcall. That's it. Get the install prompt for your AI builder →