Add a voice agent with your coding assistant

Use the audentic skill to add an existing, published OpenAI voice agent to a React, Next.js, or HTML website. It gives Codex, Claude Code, and other coding assistants the integration steps, working snippets, and checks for website permissions, microphones, and typed conversations. Create and publish the agent in audentic first; the assistant handles the website code.

Download or install

DownloadUse it for
Skill ZIPA standalone audentic/SKILL.md folder for an assistant that supports Agent Skills.
Plugin ZIPA portable Agent Plugins package with the same skill, listing metadata, and icon. Requires a compatible plugin host.
Read SKILL.mdReview the instructions or give the URL directly to your coding assistant.
Public sourceInspect the skill and plugin files or install with the skills CLI.

Both downloads contain integration instructions, not the React widget itself. They do not connect to your audentic account or grant a coding assistant permission to manage agents. The download manifest includes the release version and SHA-256 checksums. These are direct downloads from audentic; availability here does not imply approval in an OpenAI or Claude plugin directory.

Install with the skills CLI

Run this in your website repository, review the skill, and choose your coding assistant in the installer:

npx skills add detextit/audentic-agent-plugin --skill audentic

The skills CLI documentation describes assistant selection and installation scope. It is a third-party installer; you can use the standalone download instead.

Install the skill manually

Extract the skill ZIP and put its audentic folder in the relevant project directory:

  • Codex: .agents/skills/audentic/SKILL.md
  • Claude Code: .claude/skills/audentic/SKILL.md

Open a new session in the repository and ask the assistant to use the audentic skill. Hosted sessions can have different installation rules; see the current Codex skill documentation and Claude Code skill documentation. If your assistant does not support skill installation, give it the SKILL.md URL and the task below.

Prepare your agent

  1. Create a private draft from a website page, or describe your business by voice or text. The OpenAI voice agent guide walks through the full setup.
  2. Review the imported facts, add reference materials, and set Starting language in About.
  3. Test in Test & publish. Type instead lets you check answers without microphone access; also test a spoken conversation. Typed tests still use a model connection and a conversation start.
  4. Add your website's exact origin to Website addresses. https://example.com and https://www.example.com are separate entries. Include staging and the actual localhost port if you need them.
  5. Publish and copy the agent ID or generated website snippet.

An external website cannot start a conversation with an unpublished agent or a disallowed origin. The agent ID can appear in website code. You do not need to give your coding assistant an OpenAI API key: audentic manages model access and keeps credentials on its server.

Give your assistant a focused task

Replace the placeholders and paste this into the assistant working on your website:

Use the audentic skill to add our published voice agent to this website.

Agent ID: YOUR_AGENT_ID
audentic application URL: https://www.audentic.org
Website URL: YOUR_WEBSITE_URL
Placement: a floating button in the bottom-right corner

Inspect the framework, package manager, and any existing audentic integration.
Use the placement above without adding duplicate widgets. For a floating
button, use the hosted widget.js script; if I ask for an inline React card,
use @audentic/react and SessionControl in a Client Component.

Keep credentials out of the browser. Don't start a conversation automatically.
Run the project's usual build or typecheck and check the mobile layout.
Tell me the exact website origins to add in audentic, what you verified,
and which live microphone and typed-conversation checks remain.

Choose the placement you want before sending the task. The skill supports a floating button, an inline HTML card, or a React component. With repository access, the assistant can edit and check your site. In chat alone, it can provide the code and setup instructions. See the React guide or website widget guide for the underlying examples.

Check the live website

Visit the published website while signed out of audentic, including on mobile. Opening the widget should not request microphone access. Try Type instead, then separately start a voice conversation, hear a reply, interrupt, and end the call. Confirm that the microphone indicator turns off. Test the language your visitors will use.

Voice needs HTTPS or localhost, microphone permission, and a compatible browser. An enclosing iframe must allow microphone and autoplay. If your site has a Content Security Policy, follow the browser requirements and fixes. A successful build or snippet check does not verify a real conversation.

What can the skill manage?

The skill installs and troubleshoots a website integration. Configure knowledge, voices, tools, publication, and website permissions in audentic. It cannot create or publish agents through a public management API: audentic does not currently offer a supported management API, coding-agent credential, CLI, or MCP server. Visitor session tokens cannot manage agents.

See planned developer tools for future management support. The skill is published by Detextit, the operator of audentic. For help, contact Detextit or open an issue in the public repository.

Updated · Written by the audentic team