The five that come first.

Objections · answered

1. Does it hallucinate? It is hallucination-resistant by architecture, not by a better model. The model drafts freely; gates outside the model hold every line to the source it cites. What passes is rendered. What fails is held and named on the page, with the gap stated, instead of printed quietly.
2. Does client data leave the office? No. It runs on hardware in your building and ships with its own model, so the file, the questions and the output stay on your disk, under your retention policy. A frontier model is optional, off unless you turn it on, and a decision you make case by case.
3. Nobody prompts it. How do I control it? You set the boundary, not the question. It takes the material with no question attached, elects the lines of inquiry itself, and reports what it proved and what it holds. You give instructions when you have them, and every decision that belongs to a lawyer stays with the lawyer.
4. We already use a transcription or flagging tool. Why this? Keep it. Transcription turns footage into text and flags what it was built to flag. This decides, per case, what in that record bears on your issues, reasons each one through to the exhibit, and proves the conclusion or refuses it. Different job, downstream of the one you have.
5. Who is liable? You are, as you always were, and nothing here changes that. What changes is the review: every rendered line carries the exhibit, page and timestamp it came from, and whatever could not be established is listed by name. Verification becomes a spot check rather than a second draft.