gyotaku
v0.1.2Free and open source

Ctrl F for your screenshots

gyotaku reads the text in every screenshot on your computer, so you can find any of them by typing a word you remember. Free, open source and fully offline.

Install gyotaku

Linux and Windows · macOS port in review

Background: The Great Wave off Kanagawa, Hokusai, c. 1831

What it does

Everything below is live and works the way it does in the app, with the same keys.

Finds the words it misread

OCR sometimes reads an I as an l, or an O as a 0. Those screenshots still turn up, after every exact match, marked as a near match.

Try inv0ice

Fern & Co

Your invoice is readyInvoice #4021Amount due 2,340.00Due by 12 Oct
near match

Order confirmed

Thanks for shoppinglnvoice #3907Paid with UPIArrives Tuesday
near match

Mum

call me when you landok, at the gateboarding nowsafe flight!
near match
Background: Sudden Shower over Shin-Ōhashi Bridge and Atake, Hiroshige, 1857

Bursts fold into one

Six screenshots of the same chat, taken while you scrolled, show as one tile. The rest are one key away.

ctrl e unfolds them

Background: Fine Wind, Clear Morning, Hokusai, c. 1831

Copy just the lines you need

Drag a box over an open screenshot to copy the lines inside it, or click one line to copy only that.

Drag across the booking

Background: Fine Wind, Clear Morning, Hokusai, c. 1831
    • Discorderror: cannot find module 'sharp'
    • Discordbuild error on main, see the logs
    • ScreenshotsError 502 Bad Gateway
    • Discordlunch at 1? the usual place
    • ScreenshotsInvoice #4021, amount due
    • Downloadserror rate 0.2% over the week

    Filters for where and when

    in:discord, date:yesterday, before:aug. They mix with words, and dim once they're complete.

  • otp
    0 marked

    Clear out a hundred at once

    Search otp, mark every result, move them to the trash. Nothing is deleted, and Ctrl+Z puts them all back.

  • save copied images
    WHEN THE BUILD PASSES
    Clipboard12
    Pictures/Clipboard

    copy it, it gets saved and read

    Copied images count too

    Turn it on and anything you copy, even if it was never saved, lands in Pictures/Clipboard and becomes searchable. Off by default.

  • स्टेटस

    ORDERS

    ऑर्डर #4021 का स्टेटसडिलीवरी आज शाम तकपता: जोधपुर, राजस्थान
    縦書きのテキスト
    found

    Hindi, Marathi, Nepali, and vertical text

    Turn on Devanagari in settings and it reads that too. Vertical Japanese and sideways chart labels read as well.

  • SHORTCUTS

    Every shortcut is yours

    Pick a command and press the new keys. Escape cancels, Delete puts the default back, and keys already in use are refused.

Measured, not promised

On a laptop with over 5,000 real screenshots.

per keystroke, across 5,000+ screenshots
1 to 10 ms
from a screenshot saved to searchable
0.77 s
from the shortcut to the window
~120 ms
of memory while waiting in the background
37 MB
of disk per 1,000 screenshots
~30 MB
needed, it renders in software too
No GPU
How these were measured

Install in one line

No admin rights and no developer tools. Run it again any time to update.

$curl -fsSL https://raw.githubusercontent.com/xevrion/gyotaku/main/install.sh | sh

x86_64 and arm64, on any distro with glibc 2.35 or newer.

Run it again any time to update. The old window closes and the reader restarts on the new version.

Nothing leaves your computer

Screenshots hold passwords, codes, chats and bank details. gyotaku reads them on your own CPU and keeps the index on your own disk. No account, no telemetry, no update checks.

Everything gyotaku does onlineall time
setuptext reading modelonce, checksum pinned
setupONNX Runtimeonce, checksum pinned
afteryour screenshotsnever leave
afterusage datanever collected

Questions

Something else? Open an issue.

Is it really free?

Yes. gyotaku is open source under the GPL-3.0, with no paid tier, account or limits. The code is on GitHub.

Does anything leave my computer?

No. Screenshots are read on your own CPU and the index stays on your disk. The only network access is a one-time download of the text reading model and ONNX Runtime at setup, each checked against a pinned checksum. There's no telemetry and no update check.

Which systems does it run on?

Linux on x86_64 and arm64 with glibc 2.35 or newer (Ubuntu 22.04, Debian 12, Fedora, Arch, openSUSE and so on), and Windows 10 and 11. A macOS port is in review.

Does it need a GPU, or slow my computer down?

No GPU needed. Reading runs at idle priority, so it only uses time nothing else wants, and takes under a second per screenshot. Waiting in the background it uses about 37 MB of memory.

How much disk space does it take?

About 30 MB per 1,000 screenshots for the index and thumbnails, plus around 46 MB once for the model and runtime.

Which languages can it read?

English and other Latin-script languages, Chinese and Japanese out of the box, including vertical text. Hindi, Marathi and Nepali can be turned on in settings.

How do I uninstall it?

Run the install command again with --uninstall on Linux, or remove it from Settings > Apps on Windows. The README has the exact commands.