Try it free
texti
Screenshot → safe text · open source

Send your AI the text, not the screenshot.

texti reads the words in an image, hides IDs and phone numbers, sorts the result into a privacy tier, and shows how many tokens the text saves.

Processed on your device

Drop in a screenshot. Get clean text you can actually use, and a clear view of what was hidden.

Try it with a sample
Aadhaar2345 6789 0124
AadhaarXXXX XXXX 0124
Mobile98765 43210
Mobile[REDACTED:PHONE]
Tier 3 · ~765 → ~14 tokens0 uploads
0files uploaded
3privacy tiers
98%fewer tokens on a sparse ID image
159automated tests in the repo
Live tool / 01

Make a screenshot safe to share.

No account. No waitlist. No upload.

Your file

Or load a made-up sample

All samples are made up. In Image mode a sample is drawn as a picture and read with the same reader as your own files.

What your AI would get

Tier 2

    Text after masking

    As an image~765
    As texti text

    Show the text before masking

    Nothing was uploaded. The text and the image stay in this page.

    This page runs the same detectors and tier rules as the command-line tool. It handles one image or one piece of text at a time. Connecting to OneDrive, PDFs, audio and a searchable encrypted index need the command-line version below. Token counts are estimates: an image is costed at about 765 tokens, a typical figure for vision models.

    How it works / 02

    From pixels to practical.

    Everything in the dashed box runs on your computer.
    01

    Read locally

    Your browser reads English, Hindi or Telugu text from the image. The picture is never sent to texti.

    02

    Sort and mask

    Checksum and pattern detectors find IDs, cards, contacts, codes and keys. The page gets a tier, and the sensitive values are masked.

    03

    Send only the text

    Review, copy and paste the masked text. It costs far fewer tokens than the picture when the image is mostly sparse.

    Your file Read the wordsSort into a tierMask IDs and secrets Text you approveYour AI
    The privacy promise / 03

    Your files are not our business.

    This page reads the image and masks the text in your browser. There is no upload endpoint, no account and no waitlist behind it.

    1. Images stay in the current tab
    2. No text or findings are saved
    3. Closing or reloading the page clears your work

    The page loads its reader program from a public code host (cdn.jsdelivr.net) and its language data from this site. Those requests contain no image and no text.

    Privacy tiers / 04

    You decide what can leave your device.

    Unknown files start in Tier 2. Content can only raise a tier.
    Tier 1 · General

    Study notes, slides, public documents

    Returned in normal searches.

    Masked: card numbers, OTPs, keys and ID numbers.

    Tier 2 · Personal

    Photos, chats, personal notes

    Held back until you review and approve exactly what would be sent.

    Also masked: email addresses and phone numbers.

    Tier 3 · Private

    ID cards, bank statements, medical records

    Held back by default, and shown with a warning when you ask for it.

    Aadhaar, PAN, passport and account numbers always masked.

    The full version / 05

    Connect OneDrive and search everything.

    Open source. A developer tool today.
    • OneDrive sync, read-only. It can never change or delete a file.
    • Images, PDFs, audio and video. Scanned pages go through OCR and recordings are transcribed on your computer.
    • Encrypted index, one key per tier. Keys stay in your operating system's keychain.
    • Search within a token budget. Short summaries first, then the best matches.
    • Forget everything in one command. Your OneDrive files are never touched.
    git clone https://github.com/nithish-74/Texti.git cd Texti pip install -e . texti fetch-model texti doctor
    Read the README

    Needs Python 3.11 or newer, Tesseract OCR and your own Microsoft app registration. The README lists each step.

    Why bother / 06

    A screenshot versus texti.

    Attach the screenshot
    Send texti text
    Tokens per image
    About 765 (estimate)
    Often 10 to 300, depending on how much text the image holds
    Where it is read
    By the AI provider, from the image
    On your device, first
    ID and card numbers
    Visible in the image
    Masked before anything is stored or sent
    Private files
    Whatever you attach
    Held back until you approve
    Find it again later
    Scroll through your gallery
    Search within a token budget (command-line version)
    Limits / 07

    What it does not do yet.

    Good to know / 08

    Before you share.

    Is my image uploaded anywhere?

    No. The reader runs in this page. When you first read an image, your browser loads the reader's program code from a public code host (cdn.jsdelivr.net), and the language data comes from this site. Your image and its text are never sent to any server.

    Does it work for Hindi and Telugu?

    Yes, choose the language in Image mode. Accuracy is lower than for English and varies with the image, so check the confidence note.

    Can the command-line version see my files?

    Only on your own computer. It reads your OneDrive with read-only permission, processes files locally, and keeps the index encrypted on your device. The only outside connections are to Microsoft and to one-time downloads of public models.

    Does it work with ChatGPT, Claude and Gemini?

    You copy the masked text into any of them. A browser extension that does this for you is planned but not built.

    How much does it cost?

    texti is free and open source under the MIT license.

    Send your AI the text. Keep the files.

    The code is open, the limits are written down, and feedback helps decide what gets built next.

    View texti on GitHub