Skip to content
Aingel
All demos
AI agentLive AISample data

Knowledge agent

Answers from your own documents, hands off when it should, never invents prices.

Trained on five fictional docs for Harbor Home Services, a plumbing and HVAC company.

portal.harborhomeservices.com/agents

Harbor Home Services assistant

Answers from the company's own docs and gets a human involved when it should.

ReadsCompany docs
ActsBook a callbackHand off to a personOpen a ticket

Ask me about Harbor Home Services: pricing, policies, warranties, or how to do something in the field app. If you need a person, say so and I'll get one.

Ask
Do
Enter to send · Shift+Enter for a new lineReady

  • Answers from the docs

    Prices, policies, and how-tos are quoted from the knowledge base, not from the model's memory.

  • Says “I’m not sure”

    Ask something the docs don't cover and it says so and names the person to contact.

  • Refuses live-data questions

    Invoice status, schedules, orders: it has no access, says so, and never pretends it checked.

  • Knows when to get a human

    Ask for a person or describe an emergency and it hands off to a named human with a note, or books a callback with your approval. Every action is logged.

Sample data

Harbor Home Services and Coastline Coffee Roasters are fictional. Every person, price, and policy is invented. Callbacks, handoffs, and tickets here write to in-memory demo records only. Anything you paste is used only to answer this chat, so don't include confidential details.

What this mirrors

The how-to agent deployed to a client's field team.

  • Each agent answers from a curated, versioned knowledge base. A new version can be activated or rolled back without touching the agent.
  • The grounding rules are literal: answer from the knowledge base, say you're not sure otherwise, never imply access to live systems, stay in scope.
  • Persona, grounding rules, and knowledge base form one cached system prompt, so repeat questions cost a fraction of the first.
  • In the real product, users can paste a screenshot of where they're stuck and the agent reads it before answering.
  • Every answer has thumbs-up, thumbs-down, and flag controls. Flagged answers land in a review queue so the knowledge base gets fixed, not just the reply.
  • There is no hand-off button in the product: when the docs run out, the agent names the person to contact. The callback, hand-off, and ticket actions on this page show the action pattern we add when a client wants it, gated by approval and logged.

Everything on this page runs on fictional sample data. No client names, no client numbers.

Want this for your business?

Every build is scoped to your tools and your team. Book a call and we'll show you the real thing.

Book a callNext demo: Commission Tracker