An AI front desk that only says what you'd approve.
Configure the personality, the rules, and how much autonomy it gets. It answers from your knowledge, escalates what it shouldn't touch, and proves it's ready before it talks to a real customer.
How it works
You set the personality and the rules
Tone, goals, escalation keywords, and an action mode: autonomous (it sends), approval (a teammate reviews first), or off. No prompt-engineering required.
It answers only from what you approved
Every reply is matched against your structured facts and approved FAQs. No match, no guess - the question is logged as a gap for you to fill instead.
You run the readiness suite before going live
A scored set of scenarios - an unknown question, a request for a human, a prompt-injection attempt, a sensitive-domain question, and a real price question from your own services - with a pass/fail per scenario. It has to pass before the agent can answer customers on its own.
What it can do
- Grounded answers only, with the source shown
- Deterministic guardrails: prompt-injection detection and sensitive-domain refusal, checked before anything else
- Configurable escalation keywords hand off to a human instantly, and the customer is told a person is taking over
- Answering customers unsupervised is locked until the readiness suite passes
- You choose how much of a customer's history it may read, and the narrowest setting reads nothing
- Autonomous or approval-required action modes, per agent
- A playground to test messages against your live configuration without touching a real conversation
- A scored readiness suite you can re-run any time your knowledge changes
Prompt-injection detection and sensitive-domain refusal run before any answer is even looked up - a fixed, testable floor that doesn't depend on what a model decides in the moment.
A fixed set of scenarios plus a live question about your own services, scored with a pass/fail per scenario - re-run it any time your knowledge changes.
Frequently asked questions
Does the AI ever make up an answer?
No. It only answers from structured facts and FAQs you've approved. If nothing matches, it logs the question as a knowledge gap instead of guessing.
What happens with a prompt-injection attempt?
It's caught by a deterministic pattern check before anything else runs, and the conversation is handed to a human. This isn't a second AI call asked to judge the first one - it's a network-free check that can't itself be tricked by the suspicious text.
Can staff take over from the AI?
Yes, any time. In approval mode every AI reply waits for a teammate to approve or edit before it sends. In autonomous mode, any escalation keyword or sensitive topic hands the conversation to a human immediately.
Is this a chatbot with a script?
No. There's no decision tree to build. You describe the personality and rules; the agent finds the right structured fact or FAQ for each question on its own.
How do I know it's ready before customers see it?
Run the readiness suite. It scores the agent against fixed scenarios plus a real question about your own services, and shows exactly which scenario failed if the score isn't where you want it. Until it passes, the agent cannot be set to answer customers on its own: it keeps drafting replies for a teammate to send, which is where plenty of practices are happy to leave it permanently.
How much does the AI remember about a customer?
As much as you allow, and no more. Three choices: only the message it is answering, this conversation, or this customer's earlier conversations as well. A new agent starts on this conversation, which reads nothing the customer is not already looking at. The widest setting is the one you pick deliberately, and it is bounded by how far back you say.
Will it answer the same question twice?
No, when it can see the conversation. Somebody who asks a question, gets an answer and asks it again has told you the answer did not work, so repeating it is the wrong response: the agent hands the conversation to a person instead. On the narrowest memory setting it has no way to notice, and it says so on the screen.
What does a customer see when the AI hands over?
A short message saying their question has gone to your team, which is the one thing that is already true at that moment. You can write your own wording, or turn it off. It is sent in approval mode too, because somebody asking for a human should not watch an empty thread while a draft waits in your inbox.
Where to go next
Built for
Works with
Start free. Be live this week.
Built for dental, med spa, home services, and the businesses that live on the phone.