Website Chat

What the package handles

What you can build with it

• Answers grounded in approved business facts, with source references and approved resource links.
• Saved conversations that survive a service restart and a visitor-controlled delete option.
• Consented visitor messages, a private staff inbox and replies visitors can check in the same chat.
• An embeddable widget with a customizable animated assistant, reduced-motion support and an image fallback.

Supported behavior

• Exact-origin website embedding through an isolated iframe.
• Approved fact updates through a protected console, local JSON or a configured approved HTTPS feed.
• Business-specific answers, saved history, source links and an optional visitor-message handoff.

Before you choose

Will it fit your project?

Who it is for

• Businesses answering recurring product, pricing and setup questions on their website.
• Teams that want a saved visitor conversation and a private inbox for follow-up.
• Site owners who can run a small Node.js service alongside an existing website.

Runtime and connections

This package runs on Node.js 24 or later and SQLite on persistent local storage. You provide hosting and connect the package to your existing app. The included guide describes the supported integration; this is not a hosted service.

Depending on the features you enable, setup can require your own OpenAI accounts and credentials. Their usage, plans and transaction charges are separate from the source purchase. Check the supported behavior below before choosing a connection.

A useful handoff to your coding agent

How setup works

The source package includes local setup instructions, an integration guide, examples, help articles and verification resources. Your agent can use those files without signing in to a hosted reference page.

1. Check the requirements. Your agent inspects the application, runtime, database and supported connections before making changes.
2. Configure your installation. Answer the product-specific questions below, connect your own accounts where needed and retain the configuration privately.
3. Connect and verify. Use the included checks and a representative workflow, then verify any chosen external service separately before relying on it.

Questions to prepare

• Which approved product facts, prices, policies and resource links should the assistant use?
• Where will the Node.js service and persistent SQLite file run, and which exact website origins may embed it?
• Who keeps the facts current, monitors the staff inbox and sets the retention and model budget?

What your agent checks

• Run the labeled local demo without a model call, then check the embedded widget and reduced-motion or image fallback.
• Check known answers, source links and uncovered questions against the approved facts.
• Restart the service and confirm that the original conversation and request recovery remain available without another model dispatch.
• Save a consented visitor message, read it in the staff inbox and check the reply from the visitor conversation.
• Confirm stale knowledge and uncertain model outcomes stop further live replies until resolved.

Scope of this edition

Important limits

Requires Node.js 24 or later and one application process per SQLite database on persistent local storage. You provide hosting and an OpenAI account for live answers. Staff must monitor the inbox; this version does not send email or push notifications or promise an immediate human response. The assistant has no account, purchasing or other business-action tools.

• One business and one process per persistent SQLite database; no shared multi-replica deployment.
• Answers can still contain mistakes; maintain the approved facts and review representative questions.
• The daily model budget uses conservative reservations rather than a provider-enforced billing ceiling.
• No email or push notifications, immediate-human availability guarantee, account actions or purchasing tools.