It answers the questions you keep answering
Sizing, shipping, availability — answered in your voice, from your own notes, in seconds. And when it does not know, it says nothing rather than inventing something.
- Answers only from knowledge you have written — no browsing, no guessing
- Your keyword automations always go first; nobody gets two replies
- One credit per delivered reply; skips and failures are free
- Every decision logged, including the silent ones
The shape of the problem
Keyword automations are excellent at the twenty questions you predicted. They are useless for the twenty-first, which is where most of the day goes — the slightly-different phrasing, the follow-up question, the thing you never thought to write a rule for.
What the AI is allowed to do
Very little, deliberately.
It reads the message and classifies it. Greetings, small talk, emoji and abuse are left alone — only real questions, buying intent and support issues are even considered for a reply. Then it answers from the knowledge you have written and nothing else. No web browsing. No general knowledge. If your notes do not cover it, it says it does not know and offers to check, which is what a good employee does.
Your automations come first, always
Precedence is fixed and not configurable, because it should not be: a run that is waiting for a reply claims the message first, then your keyword automations get their turn, and only if neither fired is the AI considered at all. Nobody ever receives two replies to one message.
When a human takes over
Reply yourself and the AI goes quiet in that thread for 48 hours. Reply again and the clock resets. The person on the other end gets you, not a bot talking over you.
What a credit is
One credit is one reply that was actually delivered. Reading the message, classifying it, deciding to stay quiet, evaluating whether a goal succeeded, and a send that failed — all free. You are charged after it lands, never before.
The receipts
Every decision is written down, including the ones where it said nothing and why. The overview shows you the skip breakdown as prominently as the reply count, because “it stayed quiet 43 times and charged you nothing” is the number that tells you whether to trust it.
Goals, measured honestly
You can give it a goal — share a link, capture a lead, ask for a follow — and it will work towards it when the conversation allows, never at the cost of answering the actual question. Success is counted from what we can observe: the tracked link was clicked, the field was saved, the button was tapped. Never from the model saying it went well.