Most AI agents improvise the moment they run out of road. This one does not. It answers from a knowledge base written out of our own research, and when a question falls outside that, it says so and offers you a person instead. Refusing well is the hard part.
This is the live agent, not a recording. It is the same agent, on the same knowledge base, that anyone landing on this site talks to. Ask it what creative intelligence means, or how we actually work, or something we would never have thought to tell it. The interesting part is what it does when it does not know.
Your browser will ask for microphone access. The agent says up front that it is an AI and that the conversation is recorded. You can type instead of speaking.
An agent is only as honest as what you hand it. We write the knowledge base out of your own research: what you sell, how you work, who you are for, and the claims your evidence actually supports. Anything your research has not established stays out of it, which is exactly why the agent can be held to what it says. Nothing in it is invented.
A confident wrong answer costs more than no answer at all. Every question is scored against the knowledge base before the agent is allowed to speak, and anything under the line comes back as a plain admission that it does not know, followed by an offer to get a person. It will not guess a number, round a figure, or claim a credential you do not hold. The refusal is the feature.
It answers first, then it asks. The agent works out what someone wants built, how involved they want to be, and when they need it, in the order a real conversation would take it. Then it checks them against the rules you set for the work you take on, and passes the ones who fit through to you. The ones it turns away, it turns away kindly.
Every agent we build starts from one question: what would it need to know
before it is allowed to speak for you?
