Decide what the first version should handle
Pick a small set of useful questions: how to get started, where to find a feature, or how your refund policy works. Decide which requests should go to a person. Avoid presenting a documentation-based chatbot as if it can look up private orders or change accounts unless those capabilities actually exist.
Agree on a fallback message and who will monitor handoffs. An installation is not finished just because a chat bubble appears.
Create and test your agent
In Helpnest, create an agent, add your website or documents, and configure its instructions. Set a welcome message that tells visitors what it can help with. Suggested questions can make a blank chat window easier to approach.
Use the playground before installing the widget. Check your most common questions and at least one request the agent cannot answer. If your sources are not ready, follow the training guide first.
Install the widget using the supplied snippet
Open the agent’s integration settings and copy its generated script snippet. Use the snippet supplied by the app so the script URL and agent ID are correct. On a plain HTML site, place it before the closing body tag.
For a site builder, use its custom-code or footer-code area. On a React or Next.js site, follow the framework-specific installation instructions rather than adding the script repeatedly during every render. Our widget installation docs cover HTML, WordPress, Webflow, Shopify, Next.js, and Google Tag Manager.
Check the widget on a real page
Open your published site in a fresh browser session. Confirm that the bubble does not cover a cookie banner, navigation, or checkout controls. Test on a narrow phone screen as well as a desktop.
Send a question and ask to speak to a person. Verify that the conversation reaches the correct inbox. If you set allowed domains, check the exact hostname you are testing, including any www subdomain. A snippet on an unpublished editor preview may behave differently from the public page.
Launch with a review routine
Assign someone to review early conversations. Look for unanswered questions, misleading source content, and places where the visitor needed a clearer instruction. Update the knowledge rather than simply making the agent more confident.
Monitor your usage allowance as traffic grows. Model choice affects the credits consumed by AI replies, so compare the cost and answer quality using the same questions. Read models and credits before changing models across all agents.
Build on what you know.
See the setup workflow in our docs, or join the Helpnest waitlist.
Read the quickstart →