Show your agent the part of the screen that matters and say what’s wrong. It handles the rest.
Grab it, find the file, drag it in, then explain in words where the bug is. With SayFix it takes a few seconds.
“Top right, the second button, the blue one, under the price…”
Done. The agent is already fixing it.
Rough times, yours will differ
Here is what that looks like.
Select the part of the screen with the bug. Your agent sees only that, nothing else.
Tell it what’s wrong, like you would a colleague. Drop numbers to say “one”, “two”.
Type /sayfix in the chat ($sayfix in Codex). Claude Code, Codex or Cursor get it all at once, in order.
Anyone who shows a screen to an AI or to another person and is tired of explaining it in words.
Claude Code, Codex, Cursor.
“This button sticks out of the card, fix it.”
Don’t know what this block is called? You don’t need to.
“Make this one like the page next to it.”
A shot of the mockup and of what the agent built.
“The spacing isn’t like the mockup, and the font is smaller.”
Landing pages and prototypes made with AI.
“Make the price bigger and move the button above the fold.”
ChatGPT, Claude, Gemini: spreadsheets, ad accounts, CRM.
“Why is this zero? Count it by month.”
Clients, mentors, testers. SayFix is free for this, by invitation.
“First, a bigger logo; second, remove this line.”
SayFix grew out of our own product work: every feature fixes something that used to eat our time.
Select two or three places on the screen and say something about each. Speech is transcribed on your computer, offline, and English and Russian can mix in one sentence.
Don’t use a coding agent? Choose the ordinary chat in SayFix. Your marks and words land right on the picture: paste it into ChatGPT, Claude or Gemini and it reads everything.
SayFix scrolls the page to the end by itself, stitches it and cuts it into parts the agent reads without losing detail. A header that sticks to the top shows up once.
Point at a button on a website and the box snaps exactly to it. The agent gets the button’s name and its id in the page’s code, so it finds it at once. Works in browsers.
A whole screen weighs as much as seven useful pieces. Send only what matters and there’s room for hundreds of shots.
Invite a client, a designer or a tester with a link. They show it and say what to change, and you get edits your agent understands right away.
For anyone where one person builds and another reviews: studio and client, designer and developer, tester, mentor and student.
Every edit is visible to you and to whoever sent it: new, in progress, done. And the agent writes a short summary of the day.
An example. The board is coming in one of the next updates
We made sure nothing leaks that shouldn’t: not the keys in a shot, not your voice, not your team’s edits.
SayFix helps you spot them in a shot and paint them over while the picture is still yours. You confirm every find. Drag the slider.
Speech and the text in pictures are recognised right on your computer, offline. Recordings are never kept.
A package of edits is sealed on the sender’s computer. Our server only sees a sealed envelope and deletes it once it’s picked up, or after 14 days.
Shots live in your project folder, and SayFix adds it to git’s ignore list so they never land in the repository.
Pay once. Try it free for a week.
Buy it once, updates come on their own. Subscribe to hear about the launch first.
Follow the launchWe’ll post in the channel when it’s out.
Open the Telegram channel