One AI toolkit forproduct designers
The built screen to text. A critique into a change list. The words that ship polished where they already live. One menu-bar app, one shortcut.
The build drifted from the design,
and the difference is on screen
A designer's evidence is pixels. The staging build with the wrong empty state, the label nobody agreed to, the error copy that says something else than the spec. None of it is selectable, so the review becomes an hour of retyping the screen into a document. Drag a box over it and Capture gives you the exact strings, then lets you keep asking about the same screenshot. What you decide lands in a session, so the handoff you write on Thursday still knows what the second review changed.
Confluence · Team Velocity Gadget
The real Capture and ask window. Drag a box over the built screen, then keep questioning it one turn after another.
Two ways to get from a design review to filed changes
Both end with somebody changing the interface. They differ on how the words get out of the screen, and on whether the decision survives the week.
A screenshot in a comment thread
The review is the easy part. What eats the afternoon is copying an interface into prose so the people who did not attend can act on it.
- The screen is an image, so every label you want to question gets retyped by hand before you can even say what is wrong with it.
- The critique lives in three places: a call nobody minuted, a thread of pinned comments, and your own notes.
- The words that ship are reviewed last, by whoever writes the ticket, so voice and terminology drift screen by screen.
- Thursday's handoff is written from memory, because the reasoning behind Tuesday's decision was never written down anywhere you own.
Khint, on the region you drag
Drag a box over the screen and the copy comes back as text you can act on. The shape of the change list lives in a saved agent, not in your head.
- Extract text pulls the exact on-screen strings to your clipboard, and Capture and ask questions the same screenshot over several turns.
- Select your critique notes where they already sit and a saved agent returns a change list: the screen, what it says now, what it should say.
- A microcopy agent runs on the words in place, in any app, so the empty states and error messages are reviewed as one pass instead of one ticket at a time.
- With Linear or Jira connected the palette files each change and copies the key back, and no write runs before its picker has asked you.
LinearDES-142The four pieces
a design review runs on
A review is an interface you cannot select, a house shape to put the feedback in, a memory of what was already decided, and a tracker the changes have to reach. These are the four pages behind that, each a shipped feature rather than a roadmap slide.
Capture
The empty state, the error toast, the label in a build you cannot select. Drag a box and the exact strings are text again.
Agents
Your critique format and your microcopy rules as prompts with a shortcut, run on the notes and the words where they already sit.
Memory
One session per feature. The handoff you write on Thursday already knows what the second review changed on Tuesday.
Integrations
The change list becomes Linear or Jira issues, the handoff a Confluence or Notion page, each write behind an explicit confirm.
What changes across a design review
Three things, and none of them is another comment nobody reads.
- StopsRetyping an interface into a documentThe labels, the helper text and the error copy in a build are strings again in one drag, so the review quotes the screen instead of your reading of it.
- LandsThe change, in Linear or Jira, already shapedWhich screen, what it says now, what was agreed, in one format whoever filed it. One issue per run, and every write asks before it goes out.
- Gets backThe reasoning behind Tuesday's decisionThe session holds what each review settled, so the handoff spec is written from the record rather than from what you can still remember on Thursday.
Common questions from product designers
What can a product designer actually do with Khint?
Drag a box over any part of a running build and get the exact on-screen text back, or ask questions about that screenshot across several turns. Select any text (your critique notes, a copy deck, a spec paragraph), hit Cmd+Shift+K, and run a saved AI Action on it in place: turn a review into a change list, rewrite microcopy to your house voice, tighten a handoff paragraph. With Jira, Linear, Confluence or Notion connected, the same palette files the changes and publishes the handoff without you opening a browser tab.
Does Khint plug into Figma?
No, and it does not read your design files. There is no Figma plugin and no canvas integration. Khint works on two things: the region of the screen you deliberately capture, and text or documents you act on. In practice that covers the design work that happens outside the canvas anyway, which is the reviews, the copy, the specs and the tickets. If you want the frame itself in words, capture it like any other region.
Can it review the UX copy on a screen that already shipped?
Yes, and that is one of the better reasons to keep it in the menu bar. Extract text puts the exact strings from a region on your clipboard, error codes and all, so an empty state or a validation message is text you can run an Action on instead of an image you have to transcribe. Save your terminology and voice rules as an Action and run it over the strings: it comes back with what each one says now and what it should say, in the same shape every time.
Can Khint edit my spec document without wrecking the layout?
Yes, for Word, PowerPoint, Excel and PDF. An Action can take a document as its input and give back the same file edited rather than a wall of new text: the model only ever proposes replacement text for paragraphs it changed, and Khint writes those paragraphs back into the original structure, so styles, fonts, images and the rest of the file come out untouched. You get a preview of the paragraph-level diff, and the source file is never opened for writing, the result is saved as a new file.
Does my unreleased design work leave my machine?
Only what you explicitly act on. An Action sees the text you selected plus, optionally, the active session's compact summary, nothing else. A capture sends only the region you dragged, and only when you trigger it. Memory sessions live in a local SQLite database on your machine and are never synced to a server, and your Jira, Linear or Notion credentials stay in your OS keychain. Actions can also opt in to PII redaction, which swaps emails and phone numbers for placeholders before the text reaches the model.
Run your next design review with it
Free with 300 credits a month, about 10 AI actions a day. No credit card. Connect Linear or Jira when you are ready: it takes your own API token, not an admin.