Sketchnote Install

For GNOME on Linux

Circle the bug.
Say what's wrong.

Talk and draw on your screen. One Ctrl+V gives your coding agent the notes and the annotated frames.

$ curl -fsSL https://get.sketchnote.workers.dev/install | bash
View on GitHub
ActivitiesThu 10:42
Settings

Notifications

Show previews on the lock scr
Do not disturb
Sound
Cancel Save
Terminal

> - #1 (0:03) “The save button sits way off to the right.”- #2 (0:06) “And this label is cut off.”

1
2

You talk. It takes the notes.

  1. AltShiftS

    Talk and draw

    Start a capture and explain the problem out loud. Hold Shift and drag to circle what you mean, or just click on it.

  2. Every sentence lands on its mark

    Whisper transcribes your voice on your own CPU with word timings, so each sentence sits under the mark you drew while saying it. Frames are cropped to where you worked.

  3. CtrlV

    One paste, text and image

    In your agent's input, a single Ctrl+V pastes the numbered notes and then a contact sheet of the frames. It works in any app that takes pasted text and images.

Plain notes an agent can act on

Timestamped, numbered to match the frames, and short enough to read at a glance. The full bundle, with the recording and word timings, stays in a folder if the agent needs more.

- #1 (0:04) “This footer has too much text, let's cut it down.”
- (0:08) [clicked]
- #2 (0:10) “And the model name here is out of date.”
- #3 (0:11) [clicked] “This button should open the settings instead.”

Seven keys, nothing to configure

AltShiftS
Start or finish a capture
Shift + drag
Draw
Shift + click
Drop a numbered pin
CtrlZ
Undo the last mark
Esc
Cancel and discard
CtrlV
Paste notes and frames
AltShiftV
Paste the latest capture again

Your screen and voice stay on your machine.

Local by default. Transcription runs on your CPU. Nothing is uploaded; only what you paste leaves, to wherever you paste it.

Gone after a day. A capture is deleted once it has sat unchanged for 24 hours, checked hourly while Sketchnote runs. Cancelling deletes one straight away.

API if you want it. Point it at Groq, or any OpenAI-compatible server whose model returns word timings, for faster notes. Only the mic audio is sent, never the screen.

Install

Needs GNOME 50 on Wayland, an x86_64 CPU from the last decade, and ffmpeg and pw-record. The installer checks all of it and tells you what's missing.

  1. Install the binary and the GNOME extension into ~/.local:

    $ curl -fsSL https://get.sketchnote.workers.dev/install | bash

  2. Log out and back in once, so GNOME loads the extension.

  3. Download the speech model, check your mic, and choose local or API:

    $ sketchnote setup

Later, sketchnote update fetches the newest release and keeps your models.