Website chat and voice widgets

You need a published assistant and an organisation owner or administrator account. Voice also requires working speech providers and an available call line.

Create and publish

1. Open Build → Website widgets and create a widget.
2. Select a published assistant and choose chat, voice or both.
3. Add your site's exact allowed origin, for example https://www.example.com. An origin has no path; add each domain or subdomain separately. Wildcards are not supported.
4. Configure appearance, welcome text, session limits, privacy link and offered consent controls.
5. Preview the appearance, save the draft and publish the widget. The appearance preview does not run an AI conversation.
6. Copy the generated snippet into your website before the closing body tag.

<script
  src="https://YOUR-SIGSEN-HOST/sigsen-widget.js"
  data-widget="YOUR-PUBLIC-WIDGET-ID"
  defer
></script>

Use the actual snippet from your workspace. The public widget ID is designed for embedding; provider credentials stay on the server. For inline placement, use the generated container option rather than the floating launcher.

Test on your website

Load the allowed site, open the widget and send a question with a known answer. For voice, grant microphone access and check playback, interruption and End call. Voice needs HTTPS except on localhost and network access to the configured audio service.

Test action approval and refusal. If both modes are enabled, check switching between chat and voice; the server carries bounded session history across modes.

Your site's Content Security Policy must allow the script, API requests and voice connections. If a connection is blocked, use the browser console to identify the exact host and contact your website administrator.

Session and release behaviour

Each session pins the published assistant configuration; new sessions receive the latest publication. Revoking a widget or origin also blocks existing sessions. A widget allows at most three voice starts within one session, in addition to configured duration/message limits.

Public knowledge approvals and tool policies apply. Recording/analysis and persistent memory require their separate consent choices. Chat consumes model credits; paid voice also uses the USD runtime wallet.

For a custom voice interface, see the browser SDK.