Skip to main content

The product

AI Chats is three surfaces in one web application: this marketing site, an authenticated business portal, and an embeddable customer widget. They share a backend. They do not share caches, secrets, or authorization.

Create an accountSee how onboarding works

Visitor widget

You install a short script that loads a sandboxed chat frame on your website. Origin allowlisting is enforced on the public config request. The widget does not hold API keys, tenant ids chosen by the visitor, or permission to read another business's data.

Knowledge from your website

Onboarding records a public website origin. Background work discovers sitemaps and pages, fetches with SSRF protections, and prefers feeds or structured data when they exist. Unchanged pages can skip expensive reprocessing. Robots.txt is respected. Private networks and cloud metadata addresses are rejected.

Grounded conversation

The agent retrieves tenant-scoped knowledge before answering questions about your inventory, hours, or policies. Retrieved text is treated as data, not as instructions. Optional web search and maps tools stay off unless a plan explicitly enables them, and they are skipped when your own sources already answer.

Portal, takeover, and operations

Staff sign in with email and password and work in a phone-first web portal. They can list conversations, pause AI replies, assign a human takeover, and send a human message. Organizations can have several locations, employees, and roles. Billing entitlements come from Stripe webhooks on the server, not from a field in the browser.

Limits that are part of the design

  • The model is not a guarantee of factual correctness.
  • Crawl coverage of arbitrary third-party sites is not promised; blocked or partial state is shown in the portal.
  • Usage of models, tools, crawls, and index writes is metered with hard limits. Paid plans start at $99 per month, and every checkout shows the billing term before purchase.

See the onboarding steps