> ## Documentation Index
> Fetch the complete documentation index at: https://docs.inlace.co/llms.txt
> Use this file to discover all available pages before exploring further.

# Capture

> How Lace follows your work: on-device computer vision fused with the macOS accessibility tree, in any app.

Lace follows your work across the tools you already use. On-device computer vision reads the pixels, and the macOS accessibility tree supplies text, labels, roles, and interactive state. Together they give Lace a live picture of what you're working on, in any app, with nothing to integrate.

## The floating pill

The pill is a small capsule that floats on your screen. It shows what Lace is doing: dimmed when idle, a breathing glow while Lace parses, solid while a session is live, and brighter when something needs your attention. Hover to reveal controls. Press **⌘L** to show the main window if minimized.

The pill also carries the mode button. It shows whether you're in Chat or Review, and tapping it switches. Switch into Review when a moment is worth keeping, and pin a comment where it happened. See [Reviews](/concepts/reviews) for the full workflow.

## How capture works

A capture session is global — while live, any surface (chat or review) can use the live screen. Start a session with the **Start session** button in the app header, or turn on **Auto-start on review open** in Settings.

<Steps>
  <Step title="Screenshot and accessibility scan">
    Lace captures a screenshot and reads the accessibility tree: text labels, button names, input values, element roles.
  </Step>

  <Step title="Vision analysis">
    A vision model detects UI elements and identifies spatial relationships between components.
  </Step>

  <Step title="Fusion">
    Vision results and accessibility data are merged into elements with both visual position and semantic meaning.
  </Step>

  <Step title="Context injection">
    When you send a message during a live session, the parsed screen is included as structured context for that turn and can update the latest capture cache.
  </Step>
</Steps>

## What Lace can see

| Data type         | Source                      |
| ----------------- | --------------------------- |
| Text content      | Accessibility tree + vision |
| Element roles     | Accessibility tree          |
| Visual layout     | Screenshot + vision         |
| Page structure    | Both                        |
| Interactive state | Accessibility tree          |

## Caching

Lace keeps the latest parsed screen for the live session. If the active window has not meaningfully changed, Lace can reuse the cached context. When content changes, Lace captures and parses a fresh frame automatically.