Kimi Work adds long screenshots and OCR: turn screen material into a checkable list

On September 11, 2026, Kimi Work's 3.2.7 updateconnected three ways to handle material on screen: scrolling screenshots in Launcher, text extraction in the screenshot editor, and screenshots pinned as always-on-top desktop windows.

Kimi Work is the Work mode of Kimi's desktop client, accepting material and natural-language tasks. These changes happen before you hand information to AI: a long notice need not stop at the first screen, words in an image can become editable text, and the original can stay in view for comparison. Text extraction is commonly called OCR. It recognizes characters on the screen; it does not mean the notice has been fully understood.

Bring in the whole source before asking AI to organize it

Consider a community event notice with the time at the top, eligibility in the middle and fees and the deadline at the bottom. Separate screenshots can miss the ending, while copied text may have messy formatting. A long image keeps those locations in one piece of material; OCR supplies text that can be organized further.

Pinning the original helps at the checking stage. After asking AI for a list of the time, venue, eligibility and fees, you can still see the notice and check each date and qualifier. This can reduce switching windows and retyping, but phrases such as “members only” or “by Friday” still matter, and recognition may omit words.

The combination suits pages whose information is mainly text. Complex columns, very long dynamic pages or blurry small lettering can still produce incomplete captures or an incorrect reading order. Checking the input first can prevent making plans from a list that has lost a condition.

Install the desktop client through the official product entry, enter Work mode and use a version containing the update. The listed requirements are an Apple-silicon Mac running macOS 12 or later, or Windows 10 or later. The following example organizes a notice you have permission to use; adapt the request to your own material.

1. Capture the complete page. Use the global hotkey to open Launcher, the floating desktop input box. Enter its screenshot feature and select a scrolling capture. Check that the result includes the eligibility conditions and deadline from beginning to end. If the page cannot be captured completely, add the missing sections rather than asking AI to guess them.
2. Extract text and retain the image. Use OCR text extraction in the screenshot editor and pin the screenshot on top of the desktop. Compare dates, amounts, negative statements and column order, correcting text first where needed. This is input material, not a verified conclusion yet.
3. Give a specific organization task. Put the checked text in the Work input. If image context is useful, drag or paste the screenshot into Launcher. For example: “Using only this notice, list the time, venue, eligibility, fees and deadline. Include the corresponding original wording for each item. Write 'Not stated' where information is missing; do not infer it.”
4. Check and use the list. Compare each result with the pinned original. In particular, check whether “free” depends on membership and whether the registration deadline has been mistaken for the event date. Once checked, copy the list to your notes or plans.

If the original says only “Saturday at 2 p.m.,” a suitable result should retain that wording. Without context, it should not become a specific calendar date. The deliverable is a list that can be checked item by item against the original, rather than merely a shorter summary.