From a messy file to a
footed deliverable.
Every agent is the same four-stage shape you meet at sign-in: Input, Variables, Steps, Output. Here is the whole loop, start to signed-off.
Create an agent
An agent is a repeatable pipeline you build once and run every close. Start from a blank agent or clone a template that already matches a common close task.
Add your steps
Describe each step in plain language and let AI draft the transform, or write it yourself. Every edit re-runs on a sample instantly, so you watch the data change as you build - no guessing, no burning a run.
Point it at your file
Drop the messy source - a bank export, a GL, a payroll register. CSV, Excel, or tab-delimited text up to 10 MB. Use a fresh file each run, or save one the scheduled run reuses.
Run it
Answer the variables for this period (entity, date, tax basis), then watch every step execute in order. Per-column tie-outs reconcile against your tolerance; a variance stops you before a bad number ships.
Ship it, keep the trail
Download a clean, footed file plus a full run record - every step, every tie-out, every input. A reviewer verifies and signs off without ever paying for a seat.
It ends tied out, or it tells you why.
A run does not finish until the numbers foot. When they do, you get the gauge below and a record you can hand to a reviewer.