Most AI assistants improvise. This one cannot: every answer traces back to a line someone in your business signed off on, anything else goes to a human mid-conversation, and all of it leaves a record you can actually check.
Illustrative transcript. The blocked handoff is real product behavior, not a script.
Both halves are true, and most AI assistants only solve the first one.
The calls that matter most come at night, on weekends, and mid-job — from someone with a problem and a list of numbers to try. Voicemail is the competitor's lead.
Ask most assistants what something costs or how fast you can get there, and they will produce an answer. That answer becomes a promise you have to keep or walk back.
An assistant speaking for your business is your responsibility — hard to carry when nothing records what it was allowed to say, who allowed it, and when.
It gets better only when a named person decides it should.
Every response is retrieved from answers your business wrote and approved. There is no general knowledge fallback. If it is not in the library, the assistant does not know it.
Below your confidence threshold, it hands off. On a blocked topic it hands off instantly regardless of confidence. Everything the caller already said is saved on the lead, so nobody starts over.
Anything it could not answer becomes an item waiting for review, with a draft answer already written. Approving one takes about fifteen seconds on a phone.
An approved answer goes live immediately and carries who approved it and when, permanently. That record is the difference between a tool you use and a tool you can defend.
Settled facts. Hours, service area, what you do and do not take on, how you charge — instantly, from your approved wording.
Everything in between. Answered only above a threshold you set, with the handoff rate shown next to it.
Your judgment, not its own. Prices, commitments, guarantees, anything you put on the list. Confidence is not consulted. It hands off. Always.
All of it — the library, the blocked list, the threshold, who may approve — is a setting you own in the dashboard, not a promise in a brochure.
Built from the exchange automatically — the caller is never asked to fill in a form, and never sees a score, an entry number or a flag. Every line carries a name and a timestamp. The demo shows both sides side by side as you type.
Stated as what is true today. What is not yet in place is listed underneath, in the same size type.
When something needs a person — a name that matches an existing customer, a request outside what was approved, a caller who asks for a human — it is raised and put in front of staff either way. There is no status in this system a machine can set to resolved. The judgment calls stay judgment calls.
Your answers, leads and audit trail are scoped to your business, and no other customer's login can reach them. Today that runs on shared infrastructure we operate — not a separate database per customer. If a dedicated install is a requirement for you, ask, because it is not what is running now. There is no generative model in the chat answering path, so nothing a caller types is sent to a model provider or can become training data — that one is a property of how this is built, not a promise.
Per-user accounts, no shared logins, and a role split enforced in code: staff can draft, only an owner or manager can approve. Every approval, block, and settings change lands in an append-only audit log the application has no code path to edit.
Transport is TLS in front of the application. Encryption at rest, a formal backup and retention policy, a dedicated database per customer, and a signed data processing agreement are not in place today. The voice demo is the one exception to the no-model-provider line above: it uses your browser's own speech recognition, which on most browsers sends the audio to the browser vendor to be transcribed. If any of this is a gating requirement for your business, ask where it stands before you sign anything — including here.
A monthly subscription with a set number of completed intakes included. A completed intake means someone gave a name and a way to reach them — bots, wrong numbers and one-word hangups have never counted and never will. Quoted on the call against your real volume, because we would rather price the business in front of us than publish a number and negotiate away from it.
Setup and knowledge base build is a one-time item, quoted with the plan. That is the work of sitting down with your team and writing the answers the assistant is allowed to give.
Any time, in any wording. That path is never gated, delayed, or discouraged.
The honest answer is that it can only say what your business approved, which is why the approval record matters — every answer carries who signed off on it and when. The full allocation belongs in the agreement, in writing.
Chat is live today. Voice runs on the same approved library, the same blocked topics and the same approval queue, and lands in the same lead record — but it runs in the browser, not on a phone line. Taking real calls needs a telephony number and carrier setup we have not stood up yet, so there is no date to give you. Do not buy on the strength of voice until it ships.
Per topic or per answer from the dashboard, without calling support. Retire an answer and the assistant stops saying it on the next message. Your approved answer library belongs to you.
The rules engine does not know or care what business you are in — it knows what you approved and what you blocked. Rather than claim that, the demo lets you switch between three businesses and watch it: a landscaper, a law firm, and a restaurant screening its own phone. Same engine, same receipts, same refusals — only the approved answers and the blocked list change.
The demo runs against a real approved library, built for a landscaping company. Ask it something it knows, then ask what your whole yard would cost and watch what it refuses to do. Your build would hold your answers and block your topics; the behaviour is the same.