> ## 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.

# Knowledge graph

> The knowledge graph is Lace's memory: what your team saw, did, and decided, connected across sessions.

The **knowledge graph** is where everything Lace captures accumulates: screens, actions, and the comments your team pins along the way. Lace AI reasons against this record, so its answers draw on what your team has already seen and settled instead of starting from zero.

## What gets captured

<Tabs>
  <Tab title="UI states">
    A snapshot of what's on screen at a point in time.

    | Field           | What it records                      |
    | --------------- | ------------------------------------ |
    | App name        | Which application is in focus        |
    | Window title    | Title of the active window           |
    | Element summary | Detected text, buttons, inputs       |
    | Visual state    | Screenshot hash for change detection |
    | Timestamp       | When this state was observed         |
  </Tab>

  <Tab title="Interaction events">
    A transition between UI states.

    | Field                  | What it records                               |
    | ---------------------- | --------------------------------------------- |
    | Action type            | Click, scroll, keypress, navigation           |
    | Element target         | Which UI element was acted on                 |
    | User intent (inferred) | What the user was likely trying to do         |
    | Result type            | Page change, dialog open, error, or no change |
    | Discrepancy            | Whether the result matched intent             |
  </Tab>
</Tabs>

## How the graph builds over time

Pattern recognition runs against your interaction history:

1. **Transitions accumulate** into a map of how you and your team move through your work.
2. **Patterns emerge:** loops, dead ends, hot spots, escape points.
3. **Evidence is assembled** with metrics, states, and UI content.
4. **Context compounds:** Lace AI draws on the full graph when reasoning about your work in chat.

## From capture to agent execution

1. **Detect.** The graph identifies a pattern worth investigating.
2. **Reason.** You ask Lace AI about it, or Lace AI raises it during a chat grounded in a live capture.
3. **Decide.** You pin a comment in [Reviews](/concepts/reviews) and resolve it into a decision, tied to the spot where it happened.
4. **Execute.** Your agent queries the decision through [MCP](/reference/mcp) and carries it out with the full context behind it.