The appointment is forty minutes. Twenty of them are reading.

What the first pass is actually for

Not a verdict. The point of reading a summons, a lease, a benefits denial or a collection letter before the appointment is to arrive knowing three things: what kind of document it is, what it demands and by when, and which of its terms are worth arguing about. Everything after that is your judgment, and no automated read should be allowed near it.

That framing matters because it sets what a tool is allowed to claim. Main AI reads the whole document and returns the clauses that carry exposure, the dates that start a clock, and the rule each flag turns on. It does not tell you what to advise. It gets the reading out of the way so the forty minutes are spent on the part that needs a person.

What comes back, and what you do with it

1. The document type, with its own trap list. A residential lease is checked against different failure modes than an SSDI denial or a Notice CP2000. Classification is what selects the checks, so it is the first thing to confirm and the easiest to correct.
2. Flags, each quoting your document. Every risk carries the verbatim sentence it came from. If the sentence is not in the file, the flag is wrong and you discard it in seconds — which is the fastest quality check available to you.
3. The rule behind each flag. A citation to the statute or regulation the clause turns on, so you can read the section rather than the summary of it.
4. Dates as dates. Response windows, cure periods and appeal deadlines pulled out of the clauses they appear in and exported to a calendar, rather than left as prose on page nine.

How long it takes, stated plainly

One document at a time. Uploads are processed sequentially and each document is allowed up to 90 seconds before it is marked failed and the queue moves on. A morning's intake of ten documents is therefore a worst case of about fifteen minutes, unattended, while you do something else. Most finish well inside the cap; the ceiling is the number to plan around.

If a document fails — a scan too dark to read, a file that is really a photograph of a photograph — it fails on its own row with its own reason and the queue carries on. One bad scan at number seven does not cost you numbers eight through ten.

The part that stays yours

Main AI is not a law firm and does not give legal advice. It produces a reading, not a recommendation, and the reading is built to be argued with: every claim quotes your document and names a rule so that disagreeing with it is a thirty-second exercise rather than an act of faith. In an intake context that is the only property that matters, because the cost of a confident wrong answer is borne by the person across the table.