Documentation Index
Fetch the complete documentation index at: /llms.txt
Use this file to discover all available pages before exploring further.
Test the customer-visible outcome on the exact surface you intend to deploy.
| Surface | Minimum check |
|---|---|
| Quick Test | Final transcript, intended agent response, and any expected tool result |
| Script widget | Real host page, permission, microphone input, browser playback, and repeat-visit behavior |
| Iframe widget | Same checks, plus iframe permissions and the host interaction bridge when required |
| Existing Concierge Voice | Signed-in token, final transcript, signed hook, reply text, playback event, and audible response; test immediate and deferred replies if used |
| Twilio | Physical call, both spoken directions, final transcript, number binding, and external tool result if relevant |
| Python SDK tools | Dispatcher registration, one conversation with a real tool result, reconnect and duplicate-result behavior |
reply_text event does not establish that audio played, and a completed conversation does not establish that a booking or lead was saved. Keep a failed-path test: missing knowledge, backend timeout, interrupted turn, and customer hang-up.
Use conversation and turn IDs to correlate browser events, hook requests, and timeline events. Do not log identity tokens, API keys, or deployment shared secrets. Browser events.