Skip to main content

Sight & screen context

Sight lets Vanta inspect exactly what you choose on your Mac. It is an explicit capture, not continuous screen recording: select an area, one window, or all displays, then ask a question about the attached image.

Watch Sight work

Real CLI/TUI capture and visual answers · 79 seconds · no audio

The recording shows /look screen attaching a Retina capture, Vanta describing the visible design board, and a follow-up capture producing a new visual answer. The screenshot stays pending until the operator asks a question.

Use it in the CLI or TUI

CommandCaptures
/look or /look marqueeAn area you drag to select
/look windowOne window you click
/look screenEvery connected display

After the capture, Vanta shows an attachment receipt. Type the question you want answered and send the turn. The image is not submitted by the capture command itself.

/look
What is causing the error in this dialog?

Use /attachments to inspect pending context or /attachments clear to remove it before sending. You can also ask Vanta to “look at this screen”; the agent can invoke the kernel-gated look_at_screen tool when the request is unambiguous.

Use it in Desktop

Select the capture control beside the composer, then choose:

  • Select area — drag a rectangle around the relevant UI.
  • Select window — click the app window Vanta should inspect.
  • All displays — capture every connected screen.

The result appears as a removable image attachment. Add a question, then send. You can also paste multiline text or clipboard PNG, TIFF, and JPEG images directly into the composer with Cmd+V.

Privacy and boundaries

  • Capture starts only after an explicit command or capture action.
  • Vanta does not enable ambient or continuous recording by default.
  • The native screenshot file is deleted immediately after Vanta ingests it.
  • Unsent capture context expires after five minutes.
  • Receipts identify the capture source, time, scope, dimensions, and expiry.
  • The selected image follows the same routed vision-model and kernel boundary as other image context.

If the main model cannot accept images, configure a vision-capable route with VANTA_VISION_MODEL and, when needed, VANTA_VISION_PROVIDER.

macOS permission

The CLI host or Vanta Desktop needs Screen Recording permission:

  1. Open System Settings → Privacy & Security → Screen Recording.
  2. Enable the terminal application you use for the CLI, or enable Vanta for Desktop.
  3. Quit and relaunch that application.
  4. Retry the capture.

Vanta detects blank permission-denied captures and opens the correct settings pane instead of attaching an unusable image. Cancelling the area or window selector attaches and sends nothing. If a capture is too large, select a smaller area and retry.

Current proof boundary

The real CLI/TUI path has captured a Retina display and completed a visual question through Vanta. Focused tests cover area, window, all-display, cancellation, permission denial, multiple displays, Retina dimensions, oversized images, and expiry. Source and signed-package Desktop smokes pass. The remaining release proof is one successful visual question from the signed Desktop app after Screen Recording permission is granted to that app.