Our most capable AI model is now on the Scale plan. See plans

Put your documents and website to work

Your answers already exist, in PDFs, policies and pages nobody can find. Upload them or point your agent at your website, and it answers customers and staff from what your business actually says, with a sample of answers checked against your information.

What your agent does

Learns from your files and site

Upload documents or add your website. Your plan sets how many pages it learns and how often it refreshes them.

Checks its own answers

A sample of answers is checked against your information, and anything flagged comes to you to review.

Shows you the gaps

The launch check asks questions customers usually ask and shows what your agent doesn't know yet, so you can teach it.

Example conversation

Built with

LAUNCH CHECK

Tested before it talks to a customer

Accuracy

Real customer questions

The check writes realistic questions from your own information, asks your agent each one and grades the answers, so you see exactly which ones it got wrong.

Security

Attacks, before anyone else tries

It tries to trick your agent into revealing its instructions, handing over other customers' details or obeying someone pretending to be you. Nothing it asks for is actually done.

Setup

Settings that would cause problems

It reviews how your agent is set up and flags settings that could lead to wrong answers or expose data, with how to fix each one.

In a few minutes you get a score for accuracy and attacks blocked, with a fix for each problem. Run it again after every change to make sure nothing broke.

FAQ

Knowledge Management: common questions

Documents such as FAQs, policies, price lists and product sheets, plus pages from your website. You can see what the agent learned and correct anything it got wrong.

Before going live, the launch check grades its answers to realistic questions. After that, a sample of answers is checked against your information and anything flagged comes to you.