How it works
How our AI actually works
AI sounds mysterious. It is not. Here is exactly what happens when an automation runs on your business.
5 min read · updated May 16, 2026
It is a tool, not a brain
Every automation has three parts: a trigger, a model, and an action. The trigger is what wakes it up: a new lead, a new email, a new invoice. The model is the AI that reads it and decides what to do. The action is the thing it does next: drafting a reply, filing it, alerting you.
We pick the trigger and the action. The model fills in the middle.
Which model we use
Right now, most of our automations run on Claude Haiku 4.5. It is fast, cheap, and very good at sounding human. For harder reasoning tasks we step up to Claude Sonnet or GPT-class models. You do not have to pick. We do.
Where your data goes
Your data lives in your database, your CRM, or your inbox. The AI reads it, processes the request, and writes the result back. We do not train models on your data. We do not sell your data. Your data stays yours.
We log just enough to debug: request, response, latency, error. Logs are kept 30 days and then deleted.
What it is good at
Repeatable, language-heavy work. Replies, summaries, classifications, drafts, extractions. Things a human does in under five minutes but a thousand times a week.
Reading documents
One of the most useful things AI does is pull structured data out of unstructured documents. Hand it an invoice, a receipt, a signed proposal. It returns clean JSON: vendor, totals, dates, line items, all the way down. That JSON drops straight into your accounting tool, your CRM, or a spreadsheet.
No copy-pasting. No retyping. No misreading a 7 as a 1. There is a live demo of this on the demos page.
What it is bad at
High-stakes judgment calls. Anything legal, medical, or financial that needs a human signature. We build automations to draft the reply, not to send the contract.