PressTalk — one key turns speech into text

Speak. It's already text.

Any appTypes wherever you're focused
Context-awareReads the active window
Never forced to subscribeLocal / your own key / or our hosting

Scroll down

Keyboard · typed≈40wpm

01:02Backspace

Product review tomorrow at 10anm, and let's go over the design files too, I'll book the room and send round my notes from last time, and let's get the front-end folks in as well so we can settle those interaction issues.

PressTalk · spoken≈140wpm

00:18

One pass

Product review tomorrow at 10am, and let's go over the design files too, I'll book the room and send round my notes from last time, and let's get the front-end folks in as well so we can settle those interaction issues.

3.5× faster

RAW · what you actually said

Um… tomorrow, like, tomorrow at ten, let's do the product review, uh, and go through the design files while we're at it, I'll book the room and send round my notes from last time, and let's get the front-end folks in as well so we can settle those interaction issues. yeah, that's it.

↓ any app · cursor-level injection

WeChatcasual

design review tomorrow 10am — let's go over the mocks together 👍 i'll grab the room and send my notes round, can the front-end folks join too

New Messageformal

Design review — tomorrow, 10:00

We'll hold the review at 10:00 tomorrow and walk through the design files together. I'll book the room and circulate my notes from the last session beforehand. Please also invite the front-end team, so we can settle the outstanding interaction issues on the spot.

presstalkcommand

git commit -m "move review to tomorrow 10:00"

Claude Codetechnical

move the review on the schedule page to 10:00 tomorrow and link the design files

It understands you, it doesn't just transcribe you.

It can see which window your cursor is in — so you get terminology in your editor, casual in chat, formal in mail; punctuation, filler removal and speak-Chinese-write-English come along for free. And the text lands at the cursor itself, not in a transcript box.

Understandsany app · cursor-level injection

REC · push to talk

One dial. Two keys.

First word ~0.85s · streaming ~0.31s — instant once the local engine is warm.

Dial
Turn to switch mode: dictate · translate · polish.
Middle key
Hold to talk, release to type.
Bottom key
Tap for return — send it.

Press it on the device · that's your input box on the right

Any input · at the caret

Hold the middle key and talk — the text lands right here…

From the chip to the firmware, every layer is ours.

This is how the machine stacks up. Take it apart a layer at a time — not one of them is an off-the-shelf module bolted on.

An ESP32-S3 at the centre, and the whole board around it is ours — not a Raspberry Pi or a dev board.The microphone is inside the machine. Competing devices ask you to buy one separately; this one does not.Worked out, not assumed: the dial picks dictate · translate · polish; middle key talks, bottom key sends.No generic parts anywhere in the housing — every shell is tooled from our own mould, not bought in.Written from the ground up, hardware and software together — which is why every layer of this is ours to change.

Click a part to jump to it