Quick answer
When you drop a window onto Docsi AI, he takes a single image of that window through the system’s screen capture interface and reads the visible text locally with text recognition. Only what is currently visible is captured – no hidden content, no other tabs. On a Mac, this needs the “Screen Recording” permission once. If the text can be selected, copying and pasting often gives more accurate results.
Key takeaways
- Docsi takes a single image of the window and reads the text locally.
- Only the visible area is captured, no other tabs.
- On a Mac, the “Screen Recording” permission is needed once.
- Larger text and a light background improve recognition.
- Selectable text is carried over more accurately by copying.
What a window image contains
Window drop can capture the visible content of a selected window as an image. It does not contain hidden PDF pages, spreadsheet sheets or complete web pages. Local text recognition can prepare visible text from it.
For complete documents, importing the file is a better source. A link alone also does not prove that Docsi has read the linked page.
Prepare the view deliberately
- Show only the window and section the task needs. Close confidential overlays.
- Use readable zoom and bring the required lines into view.
- Pass the window to Docsi and review the captured preview and recognised text.
- Describe the task for the captured view. For additional pages, add the original file.
System permissions and missing content
On macOS, window capture requires screen recording permission. Supported gestures and system behaviour differ between Windows and macOS. If no useful preview appears, use file import and report the problem with your system version.
How window capture works technically
On Windows, Docsi uses the public Windows.Graphics.Capture interface, on a Mac Apple’s ScreenCaptureKit. Both deliver an image of exactly the chosen window. Docsi turns this image into a PNG file and reads its text with the same local text recognition used for scans – for printed text in German and English.
A single still image is created, not a continuous recording. The capture does not leave your computer; only a subsequent AI task sends the recognised text to Docsi AI. Image-capable models also receive the image when needed.
The permission on a Mac
macOS protects screen content: an app may only capture other programs’ windows once you allow it under System Settings → Privacy & Security → Screen & System Audio Recording. Docsi asks for this on the first window drop. After allowing it, you usually need to restart Docsi once.
You can revoke the permission at any time in the same place. Docsi only uses it at the moment you drop a window onto him. On Windows you choose the window deliberately; a system permission like on the Mac is not required there.
Making the view readable
Text recognition works with an image of at most 2,400 pixels per side. A very large window with small text gets scaled down and becomes harder to read. So increase the text size in the browser or program, prefer a medium-sized window and scroll to the section that matters.
Light text on a dark background, semi-transparent layers and banner ads make recognition harder. If needed, switch to a light reading mode for the capture and close pop-ups. Then check figures and names in the recognised text before building a task on it.
Private content in the window
Everything visible in the window is captured – including notifications, open chats in a sidebar or personal data in a form. Glance at the window before dropping it and close anything that does not belong in the task.
Docsi does not capture other windows, other tabs or content below the visible area. He reads no cookies, no browser history and no passwords. If you want to use a long page, take several captures or copy the text.
When copying or a file is better
A window image is handy for content that cannot be saved as a file, such as a web app or a preview. If the text can be selected, however, copy and paste is more accurate: the text arrives character by character, without recognition errors.
If the document exists as a PDF, Word or Excel file, the file itself is the best source. Docsi then reads every page with page numbers and can assign source markers precisely – with a window image that only works for the visible part.
Three examples of window drop
An online lecture with slides on screen: pause on the important slide and drop the window onto Docsi; the slide text becomes a source for flashcards. A dashboard in the browser: capture the view with the figures and ask for a short summary with numbers.
Two windows side by side, such as an assignment and your notes: drop both onto Docsi one after the other and name their roles. In the demo video on the home page, flashcards are created from two windows this way.
A worked scenario
An example: a web app at your university shows the semester’s exam dates, which can be neither downloaded nor selected. You zoom the view to 125 percent, scroll to the table and drop the browser window onto Docsi. He takes an image and recognises the text locally.
You check the recognised dates against the view, correct one shifted room and ask Docsi for a table with subject, date, time and room as a CSV file. You import it into your calendar. For the second part of the list, which was below the visible area, you repeat the capture after scrolling.
Key terms explained
- Window drop
- Dragging a window onto Docsi so he takes an image of it and reads the text.
- Windows.Graphics.Capture
- The public Windows interface for screen and window captures.
- ScreenCaptureKit
- Apple’s interface for screen and window captures on macOS.
- Screen recording permission
- The macOS permission an app needs to capture other windows.
- Visible area
- The part of a window currently shown on screen. Only this part is captured.
A worked example
A browser shows a table whose last columns lie outside the window. A window image contains only the actual visible region. Enlarge the relevant area or use a real file when the entire table is needed. An image capture is useful for visible errors or layout questions. A text source is usually better for a complete article. Also check whether notifications or other private details appear in the image.
Adapt source and scope, then use the text in Docsi. This website starts no AI task.
Check the result
If something is wrong
| Observation | Next step |
|---|---|
| Important area is clipped | Reposition and recapture the window or add the original file. |
| Screenshot treated as a full source | State the visible scope and provide a text source for additional content. |
| Capture is unavailable | Check native app, platform and window permissions. A web preview does not establish native capture availability. |
Common questions
Can Docsi read the whole web page this way?
A window image shows the captured view. Content further down, separate tabs and hidden data are not included.
Does screen capture remain active permanently?
The source is created by the specific handoff you initiate. Use the preview to review its content before a later AI task.
Can I add several windows one after another?
Yes. Each window becomes its own source; state in your request what each window stands for.
Does it work with videos?
Docsi takes a still image. Visible text in the frame, such as subtitles or slides, is read, the audio is not.
Does window drop cost credits?
The capture and text recognition do not. Credits only apply once you start an AI task afterwards.
Is the window image saved?
It becomes a source of your task in the local working store. Docsi does not upload it; only an AI task uses the recognised text.
Sources and further reading
Statements about Docsi are based on the product documentation by Impulse Apps. These primary sources explain the underlying formats, standards and methods.
- Microsoft: Screen capture (Windows.Graphics.Capture)The Windows interface Docsi uses to capture a window.
- Apple: ScreenCaptureKitApple’s interface for window captures on the Mac.
- Apple: Control access to screen recordingHow to allow or revoke screen recording for an app.