Answers from your docs, with the source
Your help centre, pricing and policies, kept fresh. Every answer shows where it came from. When there is no source, it offers a person instead of guessing.
An AI agent for your product. It answers from your help centre, recommends the right plan, and, for signed-in users, makes real changes through your own API: reschedule, update, upgrade, cancel. Every change is confirmed first. Every action is on the record.
No engineering project. Your help centre, your pricing page and your API are what it needs, and you decide what it may do.
Point it at your help centre and pricing page. Import your API. Takes about ten minutes, no engineering ticket.
Switch on the actions your users ask for most. Each one is off until you enable it, and each change needs a confirmation.
Paste one script tag. Anonymous visitors get answers and recommendations. Signed-in users get things done.
Answering questions and recommending plans are expected. Getting things done inside your product, safely, for a signed-in user, is what turns a chat widget into something people rely on.
Your help centre, pricing and policies, kept fresh. Every answer shows where it came from. When there is no source, it offers a person instead of guessing.
Upgrades, add-ons and the right tier, from your own catalog. It never quotes a price you didn't give it, and every recommendation is measured against a group that got none.
For signed-in users: real changes in their account, each one shown on a confirmation card before it happens and recorded afterwards. Visitors who are not signed in never get this.
Clinic staff, salon owners, gym managers, landlords, contractors. They don't read release notes and they don't open a ticket for a feature they already pay for. They ask.
“Where is the thing I'm already paying for?”
“I asked support and it ended up with the engineers.”
“They signed up, poked around, and we never saw where they got stuck.”
Support deflection and actions completed show within two weeks. Revenue influence needs more volume, so it is reported only once it is real.
Conversations that ended with the question answered or the task done, without a person. Visible from the first week.
Real changes made in your product through the agent, each confirmed by the user. The number a weekend demo cannot show you.
How often a conversation went to your team, and why. It falls as your docs and actions improve, and you can see which ones.
Upgrades and add-ons after a recommendation, compared with a group that got none. Reported only once it is statistically solid.
Your engineers could build a convincing demo in a weekend. We won't pretend otherwise. What follows the demo is what takes six months and never stops.
The demo will work. The question is who is still maintaining it in a year, and what your roadmap looks like meanwhile.
The AI decides what to suggest. A separate layer decides what actually happens, and it follows your settings, not the model's mood.
The AI proposes an action. A separate service checks it against what you allowed, attaches the credential, and makes the call. Your users' credentials never reach the model.
A confirmation card shows exactly what will happen. Nothing is sent until the user says yes. Cancellations and refunds stay off until you turn them on.
Import your API and every action starts switched off. You choose what the agent may do, one action at a time, from your dashboard.
Every action, every refusal and every setting change lands in an audit trail nobody can edit, including us. You read it in your dashboard.
Three plans. Every one starts as a 30-day paid pilot on your production site, credited in full toward the annual contract.
One call to agree the criteria, an afternoon to set up, a month on your real site with the numbers on your dashboard. The fee is credited toward your annual contract.
No. Every customer starts with a paid 30-day pilot on their production site, with success criteria agreed on the first call. The fee is credited in full toward the first annual contract. It keeps the pilot serious on both sides and gives you a clean way out if the numbers are not there.
About ten minutes for answers and recommendations: one script tag. Actions need a short signing step on your server so we know who is signed in. There are guides for Next.js, Rails, Laravel, Django and plain HTML, each one complete.
No. Every action from your API starts switched off. You enable them one by one, and every change your users ask for shows a confirmation card before it happens. Cancellations and refunds are off until you turn them on.
It says so and offers a person. The full transcript goes to your team by webhook or email. Unanswered questions are grouped on your dashboard so you know what to write next.
It is encrypted at rest, redacted of personal data before storage, deleted after 90 days by default or sooner if you ask, and never used to train models. Where it is hosted and every provider that touches it are on the security page. A data processing agreement and the sub-processor list come with every pilot.