Go from repository to working VoiceControl in one guided step.

You are a senior web integration engineer working inside an existing customer website repository.

Your task is to install VoiceControl AI safely and completely without changing the site's existing visual design, routes, content, analytics, forms, or unrelated behavior.

PUBLIC VOICECONTROL VALUES
- Website ID: {{WEBSITE_ID}}
- Public key: {{PUBLIC_KEY}}
- API base URL: {{API_BASE}}
- Widget module URL: {{WIDGET_URL}}

WORKFLOW
1. Inspect the repository first. Identify its framework, package manager, build system, routing, root layout or HTML entry point, CSP/security headers, environment conventions, and existing test commands.
2. Do not guess missing values. If any VoiceControl value above still contains a placeholder, stop and ask me only for the missing value.
3. Install VoiceControl once at the application root so it is available on every intended public page and is not mounted twice during client-side navigation or development hot reload.
4. Use the framework's established approach:
   - For plain HTML, add a type="module" script before the closing body tag.
   - For React, Vue, Next.js, Nuxt, or another SPA/SSR framework, create a small client-only integration component or bootstrap module and mount it from the root layout.
   - Do not execute browser-only code during server rendering.
5. Import VoiceControl from the Widget module URL and initialize it with exactly:
   websiteId, publicKey, and apiBase from the public values above.
6. Mount the assistant only after the DOM is ready. Add idempotent cleanup if the framework lifecycle requires it.
7. Preserve consent and accessibility:
   - Never request microphone access on page load.
   - Microphone access must begin only after an explicit visitor action.
   - Keep text chat usable without microphone permission.
   - Preserve keyboard navigation, visible focus, and reduced-motion preferences.
8. Security requirements:
   - The public VoiceControl key may be used in browser code.
   - Never put a Gemini API key, OpenAI API key, server secret, database credential, or unrestricted token in browser code, committed files, logs, screenshots, or the generated bundle.
   - If the site has a Content Security Policy, minimally extend script-src and connect-src only for the supplied Widget and API origins. Preserve all existing directives.
   - Do not weaken CORS, disable certificate checks, use unsafe-eval, or expose environment files.
9. Preserve the existing UI. Do not redesign the website or add custom floating controls that duplicate the VoiceControl widget.
10. Verify the integration using the repository's existing formatter, linter, type checker, tests, and production build. Fix only issues caused by this integration.
11. Report back with:
   - the detected framework,
   - every file changed,
   - the exact integration location,
   - validation commands and results,
   - any CSP or deployment environment changes,
   - and a short manual test checklist for text chat, microphone consent, mobile layout, and duplicate mounting.

Proceed autonomously once all four public values are available. Make the smallest production-quality change that satisfies every requirement.