Nocturne Lens
Reads your screen. Or just tells you the gist.
Drag a floating frame over anything — a chat, an article, a terminal, a game — and Nocturne Lens reads it aloud as new text appears. Or switch on Summarize and it does the opposite: it watches quietly, waits for the text to stop moving, and tells you in a few spoken sentences what just happened. It never opens a file, never asks for a login, and never sends what it sees anywhere.
Windows today, macOS planned · No card, no account · Trial and checkout open soon
The trial installer is being packaged — it will be available right here at launch.
Text you cannot open, read anyway.
Most read-aloud tools need a file. Lens needs a rectangle. Put it over a group chat, a support ticket, a wiki nobody exports, a stat block in a game, subtitles in a video player — if the pixels show text, Lens can speak it.
It reads what passes through the frame rather than a document with a beginning and an end, so there is no import step and nothing to convert. You drag it, you press the moon, it starts talking.
The inside of the frame is a hole: everything under it stays scrollable and clickable, and only the ring, the moon and the pills take the mouse. It remembers its size and place, tells you when it’s narrower than the text, and a speed pill runs from 0.85× to 1.6×.
What appears on your screen is read on your machine. It is never uploaded, and no transcript of it is saved.
A rectangle, not an import dialog.
When it is more text than you want to read.
Switch on Summarize and Lens stops reading along. It collects quietly while text streams past, waits until the stream goes still — quiet, not a stopwatch, so it never cuts a thought in half — then speaks a few sentences telling you what it just watched.
Point it at an AI coding agent working through a problem, a build that is still running, a log, a thread that got away from you. You get the shape of what happened without reading a novel of it, and you can keep working while it talks. It keeps watching the whole time it is summarizing, so nothing slips past unseen.
Three lengths, cycled by right-clicking the Summarize pill: a sentence or two, a short spoken paragraph, or a full pass through the main points in the order they appeared. The summary is written by the model on your own machine.
The text it summarizes is never written down — not to a file, not to a log.
A wall of output in. Three sentences out.
It keeps up, and it does not repeat itself.
Screens move. Lens waits for a line to settle before it speaks, so half-rendered text never gets read aloud, and it tracks text that merely scrolled upward so the same sentence is not read twice.
Wrapped lines are rejoined into whole sentences before anything is spoken, text the frame edge slices through is dropped rather than guessed at, and a pass that is mostly sliced words is thrown away rather than voiced as garbage.
All of that reasoning happens in memory, on your machine, and is discarded as it goes.
Spoken once, when the line has stopped moving.
Pin it to the window doing the work.
Clamp the lens to the window underneath and it follows that window — moving with it, scaling when it resizes, and going quiet while something covers it. Set it over the panel your agent writes into and it stays there while you work in another app.
Or leave it floating, and it announces each new window that slides beneath it, so you know what it started reading without looking — a pill on the frame turns that off.
Everything it collects lives in memory, and the lens saves no transcript of what it read. None of it is uploaded.
Stays on the window that is doing the work.
Talk to it. It talks back.
Hold the Lens key and speak — what you say is a command, never typed anywhere. “Bigger.” “Fit the whole article.” “Summarize.” “Read it in full.” “Faster.” “Again.” “Follow this window.” “Put your window over the article in my browser.” It moves, resizes, clamps and answers in its own voice. You don’t even have to open it first: a command spoken to a closed Lens opens it.
Ask what it sees — “what’s on screen?” — and it tells you how many blocks of text it found and what the largest says. Say “screenshot this into the chat window” and it grabs exactly what the frame shows and pastes it where you named. Send it to another monitor: “the second screen.”
It can click things you name — “click Send”, “open Settings” — but only if you switch that on. It always says what it’s about to click, asks first before anything that looks permanent, and never touches a sign-in, password or payment field.
Prefer no key at all? Set a wake word: “Nocturne view…” goes to the Lens, “Nocturne speak…” is typed. Say “stop” and it stops, whatever else was in the sentence.
Spoken commands are understood on your machine — never typed into anything, never sent anywhere.
Say where to look and what to do. It says what it’s doing.
The Pledge
A screen reader that keeps your screen to itself.
No account
Install and open. That is it.
No cloud
The text recognition and the voices are on your disk.
Nothing uploaded
What the lens reads is recognised on your machine and never sent anywhere. The reading feed is never written to disk.
Works offline
Pull the cable and nothing changes.
Everything on your screen, out loud.
Nocturne Lens will be $59, one-time — or $149 for all four. Checkout and the free trial open soon.
See the suite