---
title: "Clips: Talk to the Agent"
description: "What the agent sees when you're in Clips, prompts organized by task, and how it reaches for less common actions like Rewind's screen memory."
search: "Clips agent prompts talk to the agent context screen awareness"
---

# Talk to the Agent

Everything in Clips is available to the agent through the same actions the UI uses, so anything you can click, you can also ask for in plain English. This page covers how the agent sees your screen, useful prompts by task, and where it reaches for less common actions. See [Clips](/docs/template-clips) for the overview.

## How the agent sees your screen

With a recording open, the agent knows exactly which one, down to the side panel you're looking at (comments, transcript, or insights) and the timestamp you're paused on. That's why "add a comment here" or "trim this to the good part" resolves correctly without you naming the recording or the time.

Browsing the library instead, the agent knows what you're looking at there too: the current folder, your search, and which recordings you've selected if you're in selection mode.

One thing the agent can't do on your behalf: start a recording. Browsers only grant microphone and screen access after a person clicks something, so "record my screen" takes you straight to the recorder instead of starting capture invisibly in the background.

## Useful prompts by task

Every prompt below becomes a real action against your actual library, the same one the UI reads and writes. Here's one worked example:

<Comparison>

### :blue: You ask

"Summarize this clip for a product update."

### :green: What happens

This is a read-only question. The agent reads the transcript for the recording you have open and answers directly in chat. Nothing on your screen changes, since nothing was created or moved.

</Comparison>

<Cards columns={2}>

### Capture and import

- **"Record my screen."** The agent opens the recorder for you. Starting the capture itself is still your click, since the browser requires it.
- **"Import this Loom link."** Paste the URL and the recording joins your library once the import finishes.

### Clean up and organize

- **"Regenerate the title and summary."** The agent rewrites both from the current transcript.
- **"Add chapters. This one's long."** Chapter markers appear on the timeline right away.
- **"Fix the mis-transcribed word at 1:42."** The agent corrects the transcript in place.
- **"Find the clip where we discussed the rollout plan."** A read-only question. The agent searches across every transcript in your library and answers in chat.

### Edit

_With a recording open, "this" resolves to the clip you're looking at._

- **"Trim the first 10 seconds."** The cut appears on the timeline immediately, and nothing is re-encoded until you export.
- **"Remove the ums and uhs and re-stitch."** The agent finds the filler words in the transcript and applies the same kind of trim.
- **"Combine these three clips into one, in this order."** The agent stitches them together, preserving the transcript and chapters.

### Share and review

- **"Who's watching this clip?"** A read-only question. The agent answers with view counts, split between people and other agents reading the transcript.
- **"Share this with the design team, view only."** The recording's visibility and access update immediately.

</Cards>

## Less common actions

Not every Clips capability is a first-class action the agent has preloaded. Rewind's screen-memory actions are a good example, since they only matter once you've turned Rewind on. When a request needs one of these, the agent searches for the right tool by name before calling it, the same way you might search a command palette instead of memorizing every shortcut. You'll typically notice this as a brief pause on requests that go beyond everyday library and editing tasks.

## What's next

- [**Clips**](/docs/template-clips) — the overview and what it replaces
- [**Features**](/docs/template-clips-features) — every feature area, with setup steps for each
- [**Clips in a Multi-App Workspace**](/docs/template-clips-integrations) — agent-readable clips, Brain export, and CRM call evidence
- [**Developer Guide**](/docs/template-clips-developers) — every action, the data model, and how to extend the app
- [**Context Awareness**](/docs/context-awareness) — how the agent knows the current clip and playhead