Seven kinds of caseload
Each of these has its own page, written to the documents that actually come through that door.
Or by the task, not the role
The same work, described by what you are doing rather than who you are.
What it does, stated as plainly as we can
- It reads the whole document, not a summary of it — PDFs, phone photos of a letter, scanned pages, or pasted text.
- Every flag quotes your document and names the rule it turns on, so the first thing you can do with a finding is check it.
- Dates come out as dates. Notice windows, response deadlines and cancellation cut-offs become entries you can export to a calendar.
- It drafts the reply — the letter, the appeal, the counter-offer — with the clause quoted and the rule named. Nothing is sent for you.
How long a document takes, and why that matters at volume
One document at a time. Main AI processes an upload queue sequentially — the next document does not start until the one before it has finished or failed — and each document is allowed up to 90 seconds before it is marked failed and the queue moves on.
| Documents | Worst case, start to finish |
|---|---|
| 1 | up to 90 seconds |
| 5 | up to about 7 ½ minutes |
| 10 | up to about 15 minutes |
| 20 | up to about 30 minutes |
Those are ceilings, not averages — most documents finish well inside the cap. But the ceiling is the number to plan a morning around, so it is the number printed here. Nothing about this runs in parallel today.
A document that fails is marked failed with its own reason and the queue carries on to the next one. A bad scan at number seven does not stop numbers eight through twenty.
What it is not
Main AI is not a law firm and does not give legal advice, and it is not a substitute for the judgment you were hired for. It is a first read: it tells you what the document says, where the exposure is, and which rule each flag turns on. Deciding what to do about that is your job, and the analysis is built to be argued with rather than trusted.
We also do not publish customer names, counts, testimonials or usage statistics, because we have none we could source. What we offer instead is that every claim in an analysis is checkable against your own document — here is exactly how to check one.
Try it on a document you already know the answer to
That is the honest way to evaluate this: run something you have already read closely, and see whether the flags match what you found.
Read a document →