Tier 2 · Safe

Honest answers from your own verified claims.

An Agent on your site that answers customer questions. Not a scripted pop-up, not a generic chat widget, not a made-up answer machine. It reads from the same verified claims file that built your site, and it never invents what it does not know.

I · How it works

Four steps, one rule: no hallucination

  1. You approve a claims file. Hours, services, prices, policies. Nothing lands on the site or in the Agent's mouth until you sign it.
  2. The Agent reads from that file, and only that file. If a visitor asks something the file does not cover, the Agent says so and offers to take a message.
  3. Booking and lead capture are available. Agent can hold a calendar, send an SMS when someone wants to be called, or hand off to your existing booking tool.
  4. Workflow configurations on request. If your operation needs the Agent to do more than answer (intake forms, triage, appointment shaping), that lives inside Tier 2. Not a separate product, not a separate price tier.
II · The pillar

Why this is the "Safe" pillar page

The Agent never hallucinates, because it is not allowed to. Every answer traces to a claim you approved. If the claim is not there, the Agent does not guess. It does the safe thing: offers to pass a message to a person. That is the only way a small business can put an AI in front of customers without risking an invented price, an invented policy, or an invented promise. "I do not know" is a feature, not a failure.

III · Vocabulary

Why we say Agent, not chatbot

Chatbots have a history of scripted, broken, and frustrating answers. The Agent is a different category. It is built, configured, and trained on your actual claims file. Some BeargrassAI client sites surface the Agent as "Assistant" when their own brand register calls for softer language. Never "chatbot," never "bot," never a cute name. The Agent stands on its own.

The pattern is already running on a BeargrassAI client site

Building an Agent for your business is not a new engineering problem for us. The same claims-first, retrieval-constrained pattern is live on a SageTechTN client site today: Cloudflare Worker for the runtime, D1 for the knowledge base, Workers AI for the language model, strict allowlists, and a small widget on the page. It refuses to answer what it cannot verify, captures a lead when the visitor asks for one, and has been through a full QA cycle. When we quote a Tier 2 build, we are not prototyping; we are deploying a pattern that already works.

Talk to Mark about adding an Agent

The Agent requires The Website as its foundation. Start there if you have not already.