The lens · one pass · ~30 seconds

Does your AI
ever tell
you no?

Paste a conversation and give it thirty seconds. You'll get every message marked — where it held, where it wobbled, and the exact message where it caved. We don't keep anything you paste.

This is the lens — one reader, one pass, a quick look. Verdicts are a different thing: they live in the docket, with three seeded runs, two blinded judge families, and a paper trail. A lens shows you where to look; it doesn't bang a gavel.

The only one being tested here is the AI — never you.

Reader: pinned at the lens API — that's the requested reader model; the provider may still route or re-version it. The reader prompt is public in the repository.

    Before the read

    A quick word first. We can't verify authenticity — a read can reflect fabricated input. And what you get is a single automated read — not advice, not a verdict, not a diagnosis, and never one of the docket's rulings.

    Your conversation is processed in memory and then discarded — not stored, not logged, not used for training. It makes exactly one trip: to DMXAPI, the third-party gateway that serves the reader model, under that provider's data policy. More detail: data handling.

    This tool is not for conversations involving harm. If you or anyone in the chat is in danger or crisis, skip the screenshots and reach a human — findahelpline.com lists free, confidential helplines by country.

    镜读也能读中文对话——直接粘贴即可,同样一遍读完、同样不存储。中文界面正在路上。

    Testing your own product? The runner is the real procedure